Physica Medica 113 (2023) 102653 Contents lists available at ScienceDirect Physica Medica journal homepage: www.elsevier.com/locate/ejmp Africa’s readiness for artificial intelligence in clinical radiotherapy delivery: Medical physicists to lead the way Eric Naab Manson a, Francis Hasford b, Chris Trauernicht c, Taofeeq Abdallah Ige d, Stephen Inkoom b, Samuel Inyang e, Odette Samba f, Nadia Khelassi-Toutaoui g, Graeme Lazarus h, Edem Kwabla Sosu i, Mark Pokoo-Aikins b, Magdalena Stoeva j,* a University for Development Studies, Tamale, Ghana b Ghana Atomic Energy Commission, Accra, Ghana c Stellenbosch University, Cape Town, South Africa d University of Abuja, Abuja, Nigeria e University of Calabar, Calabar, Nigeria f General Hospital of Yaoundé and University of Yaoundé I, Cameroon g Centre de Recherch Nucléaire d’Alger, Algeria h Inkosi Albert Luthuli Central Hospital, Durban, South Africa i School of Nuclear and Allied Sciences, University of Ghana, Ghana j Medical University of Plovdiv, Plovdiv, Bulgaria A R T I C L E I N F O A B S T R A C T Keywords: Background: There have been several proposals by researchers for the introduction of Artificial Intelligence (AI) Artificial intelligence technology due to its promising role in radiotherapy practice. However, prior to the introduction of the tech- Radiotherapy nology, there are certain general recommendations that must be achieved. Also, the current challenges of AI must Machine learning be addressed. In this review, we assess how Africa is prepared for the integration of AI technology into radio- Deep learning Quality assurance therapy service delivery. Methods: To assess the readiness of Africa for integration of AI in radiotherapy services delivery, a narrative review of the available literature from PubMed, Science Direct, Google Scholar, and Scopus was conducted in the English language using search terms such as Artificial Intelligence, Radiotherapy in Africa, Machine Learning, Deep Learning, and Quality Assurance. Results: We identified a number of issues that could limit the successful integration of AI technology into radiotherapy practice. The major issues include insufficient data for training and validation of AI models, lack of educational curriculum for AI radiotherapy-related courses, no/limited AI teaching professionals, funding, and lack of AI technology and resources. Solutions identified to facilitate smooth implementation of the technology into radiotherapy practices within the region include: creating an accessible national data bank, integrating AI radiotherapy training programs into Africa’s educational curriculum, investing in AI technology and resources such as electronic health records and cloud storage, and creation of legal laws and policies to support the use of the technology. These identified solutions need to be implemented on the background of creating awareness among health workers within the radiotherapy space. Conclusion: The challenges identified in this review are common among all the geographical regions in the Af- rican continent. Therefore, all institutions offering radiotherapy education and training programs, management of the medical centers for radiotherapy and oncology, national and regional professional bodies for medical physics, ministries of health, governments, and relevant stakeholders must take keen interest and work together to achieve this goal. * Corresponding author. E-mail addresses: haspee@yahoo.co.uk (F. Hasford), cjt@sun.ac.za (C. Trauernicht), nosambacm@yahoo.fr (O. Samba), nadiakhelassi@yahoo.fr (N. Khelassi- Toutaoui), graeme.lazarus@ialch.co.zai (G. Lazarus), eksosu@ug.edu.ghj (E.K. Sosu), ms_stoeva@yahoo.com (M. Stoeva). https://doi.org/10.1016/j.ejmp.2023.102653 Received 12 May 2023; Received in revised form 30 July 2023; Accepted 5 August 2023 Available online 14 August 2023 1120-1797/© 2023 Associazione Italiana di Fisica Medica e Sanitaria. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 1. Introduction methods, identify issues with QA measurement, and establish proac- tive measures [14]. The absence of these requirements could limit or Radiotherapy involves the use of ionizing radiation (e.g., photons, prevent the full implementation of the technology into clinical practice. electrons, and protons) beams to treat cancer. A series of procedures In order to assess the readiness of Africa for the integration of AI tech- make up the therapy process including; disease diagnosis, dose pre- nology into radiotherapy practice, we will highlight the current status scription, treatment simulation, planning, plan review, beam delivery, and challenges facing radiotherapy services in Africa, discuss the ap- and patient follow-up [1]. Nearly 50% of cancer patients worldwide plications of AI technology in radiotherapy, and how AI would improve require radiotherapy at some point during their cancer journey, making radiotherapy services in Africa. Similar analysis has been carried out for it an essential part of cancer treatment [2]. There are various machine medical imaging, which is closely related to radiotherapy [15]. types that can be used to deliver radiotherapy, including brachytherapy equipment, kilovoltage machines, and external beam high-energy radi- 2. Radiotherapy services in Africa ation machines [3]. When combined with surgery or chemotherapy, radiotherapy can both treat and eliminate tumors; additionally, it can The development of radiotherapy began following the discovery of x- ease the suffering of people with terminal cancer (i.e., palliation) [4,5]. rays in 1895 and radioactivity in 1896. A few months after its discovery, However, radiotherapy workflow involves many complex tasks, X-rays were used for the first time to treat cancer [6]. Megavoltage including tumor and organ segmentation, dose optimization, outcome therapy and brachytherapy standard operating procedures were devel- prediction, and quality assurance (QA), which can be very complex and oped through scientific advancements, trial and error, and technological time-consuming potentially affecting the quality of treatment outcomes developments. With the introduction of the Cobalt-60 machine and the [6,7] coupled with the global rising burden of cancer [8]. For instance, medical linear accelerator in the 1950 s, megavoltage therapy reached skin motion in relation to internal anatomy can limit reproducibility and its peak [17]. induce systematic setup errors. Also, the process of organ delineation Africa’s involvement with radiotherapy dates back to 1929 in North may further induce a systematic error during treatment planning [9]. Africa when Morocco opened the Bergonié Center at Casablanca’s Moreover, patient-specific quality assurance requires a lot of time, Averroes Hospital. Later in 1930, Egypt’s Kasr Al-Ainy Hospital (Cairo which may result in machine downtime and interfere with patient care University Hospital) built its first radiotherapy department [16]. In [10]. North Africa, 145 megavoltage radiotherapy machines (MVM) across 85 Innovations in treatment technology are desperately needed given centers have been documented. Of the total, Cobalt-60 units make up the rising incidence of cancer worldwide and the significant discrep- 31% (45 units) whiles 69% (100 machines) are linear accelerators. In ancies in radiotherapy services [11]. Artificial intelligence (AI) tech- addition, to the megavoltage machines (MVMs), there are 36 brachy- nology has been suggested as a method to improve the standardization, therapy after-loading devices in the region. As of February 2012, there speed, and quality of radiotherapy workflow, ultimately resulting in were 74 MVMs in Egypt, 30 in Morocco, 19 in Algeria, 15 in Tunisia, and more precise and safe radiation administration through automation. 7 in Libya. The population that one MV machine can serve varies be- Machine learning (ML) and deep learning (DL) are subdomains of AI, tween these countries. For example, in Tunisia and Algeria, there were which is defined as a set of algorithms that execute tasks connected with 1.36 and 0.54 MV machines per million respectively. Also, in Egypt, human thinking or intelligence [6]. The tasks that typically require Libya, and Morocco, there were 0.9, 1.29, and 0.9 MV machines per human intelligence, such as visual perception, pattern recognition, million respectively [4]. decision-making, and problem-solving, are carried out at a similar or In 1969, the Lagos University Teaching Hospital (LUTH), Nigeria higher level of performance by means of the development and applica- became the first facility in West Africa to purchase a cobalt-60 tele- tion of complex computer algorithms [7]. Deep learning has the ability therapy machine [18]. Later in the mid-1970 s, MVMs were also to automatically extract features from huge amounts of data. Addition- commissioned in Liberia, Senegal in 1989, Ghana in 1997, Mauritania in ally, deep learning may uncover details in images that the human eye is 2010, Mali in 2012, and Côte d’Ivoire in 2018, while the rest of the West unable to see, which is crucial for the early identification of malig- African countries had no history of external beam radiotherapy services nancies using image data [12]. On the other hand, machine learning [19]. Over the past five decades, Western African countries have seen a focuses on developing predictions by utilizing mathematical algorithms surge in the total number of MVMs and per-capita radiation capacity. As to find patterns in data. For instance, automated actions can be per- of 2019, Nigeria recorded the highest number (i.e., nine) of the 22 formed by machine learning algorithms using patient data to help in the MVMs followed by Ghana (five), Senegal (three), Cote d’Ivoire (two), identification and diagnosis of cancer [13]. Mauritania (two), and Mali (one) [19]. It is recommended that commissioning, clinical implementation, and Also, the beginning of radiotherapy-assisted cancer treatment in East then daily use of the AI models together with model- and case-specific Africa was reported in the 1970 s when a Cobalt-60 EBRT treatment Quality Assurance (QA) are in place before successfully integrating AI system was supplied to Kenyatta National Hospital by the Karolinska technology into clinical practice [6]. The commissioning process uses a Institute as part of a Swedish and Kenyan government collaboration large amount of data to train an AI model, after which the accuracy of [20]. In central Africa, work on a radiation facility in Zambia’s capital the model and reproducibility is evaluated, verified, and implemented. city of Lusaka’s University Teaching Hospital began in 2003. Zambia’s To guarantee the safe and clinically appropriate use of a model, a 13 million residents did not have access to radiotherapy until 2006 [21]. multidisciplinary team who have adequate knowledge of machine Also, the plan for the first modern radiotherapy cancer center in Rwanda learning and deep learning models must be engaged prior to clinical was established in 2016 and later became operationalized in 2019 [22]. implementation [6]. This implies that all staff participating in radio- By 1994, one Cobalt-60 machine and three linear accelerators, and five therapy processes, such as radiation oncologists, medical physicists, high-dose-rate after-loading brachytherapy systems were distributed radiation therapists, IT specialists, etc., must have completed some sort across South Africa. Twenty thousand cases were treated annually by of basic education or training in AI machine learning and deep learning fifty-eight therapists and one hundred and ninety therapy radiographers techniques. [23]. Each member of the radiotherapy professional team, notably the The population served by each MVM is a critical factor in deciding medical physicist, must develop an attitude of knowledge sharing and the radiation services provided by countries. 1 MVM is what the IAEA collaboration in the application of AI in radiotherapy practice. This advises for every 250,000 people. None of the 54 nations in Africa is able would increase the capacity of medical physicists to be able to evaluate to satisfy this recommendation. Even Mauritius which is known to have the strengths and weaknesses of AI technology, produce high-quality the lowest population-to-MVM has failed to satisfy this recommendation automatic treatment plans using machine and deep learning-based [24]. In 2015, with an estimated 1.2 billion people, Africa only 2 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 possessed 394 teletherapy units. Of those, 376 were megavoltage units. linear accelerators are used in 11 (20%) countries, and only cobalt-60 There were 3 million patients per treatment unit that had access to units are used in three (11%) [28,30]. In 2030, it is anticipated that radiotherapy technology [25]. External beam (EB) radiotherapy was between 1500 and 2000 MV units and about 350 brachytherapy after- accessible in 25 of the 54 countries in Africa. The majority of the MV loaders will be required, depending on the pace of radiation use and capacity (i.e., 57%, or 209/364) was located in Egypt and South Africa patient throughput [26]. Table 1 shows details of current data on [26]. While LINAC-based radiation therapy facilities are available in 28 radiotherapy resources for patient treatment in 33 African Countries African nations, there are sadly none at all in the remaining 27 nations. [31]. South Africa (97) and the Mediterranean nations (227) are where you can find the majority of LINACs in Africa. In the 28 nations having 3. Current challenges of radiotherapy delivery in Africa LINAC-based radiation therapy facilities, there is one machine each for every 423 000 people, nearly 5 million people in Kenya, and more than Table 2 shows a summary of the main challenges faced by radio- 100 million people in Ethiopia. One radiation therapy equipment is used therapy facilities in Africa. It is challenging for cancer patients to receive for every 87,000, 119,000, 134,000, and 195,000 individuals in HICs the care they require in Africa due to inadequate radiotherapy equip- including the USA, Switzerland, Canada, and the UK respectively. The ment, limited human resources, insufficient education/training of number of machines to people in Africa is one LINAC per 3,000,000 radiotherapy personnel, and lack of innovation in treatment technology people [27]. In middle Africa with a total of 100 machines, the number [11]. As of 2014, Senegal, for instance, had 3 Medical Physicists and 2 of radiotherapy machines per million people is 0.078, which is below the Radiation Oncologists. This radiation team sees 50 patients each day, of recommendation of 1 [28]. which 40% have cervical cancer [29]. Also, Ethiopia currently has only On the other hand, less than half of the African countries (i.e., 22 of one operational cobalt teletherapy equipment that serves more than 100 the 54) have access to brachytherapy services [29]. Cervical cancer, a million people and treats more than 1,700 radiotherapy patients annu- tumor with a very high prevalence in the continent of Africa, is mostly ally [32]. According to Stefan’s investigation of the resources available treated with brachytherapy as an intracavitary procedure. Due to for childhood cancer in Africa, there are several difficulties in providing reduced treatment times, high-dose-rate brachytherapy (HDR) is rec- care for children with cancer. These include a lack of clear therapy ommended for facilities treating a large number of patients. Although protocols, inadequate radiation facilities and staff, poor personnel ed- Sievert first proposed the idea of remote after-loading in 1937 to reduce ucation or training, and a lack of research time [33]. radiation exposure to workers, a couple of countries, mainly in sub- African nations with zero Oncologists include Lesotho, Benin, Saharan Africa, still use manual after-loading techniques today. Africa Gambia, South Sudan, and Sierra Leone. On the contrary, Malawi, is yet to have access to electronic brachytherapy [25]. Fig. 1 shows a Burkina Faso, Rwanda, and Togo each have one oncologist, and Egypt map comparing the total number of equipment (i.e., radiotherapy (RT) has up to 1,500. This results in an extremely high caseload for each and brachytherapy (BT)) per million population in Africa, Western radiation therapist [34]. Sub-Saharan Africa needs at least 2000 radia- Europe and North America. tion oncologists and 1250 medical physicists, which is almost ten times As of March 2020, there were 430 megavoltage units in total scat- more than what is already available, due to the high cancer prevalence tered unevenly over the African continent with the majority in the [4]. According to the IAEA, a basic radiotherapy clinic with one MV unit Northern and Southern parts. Of the 54 African countries, 28 (52%) had should contain seven radiotherapy treatment teams, three to four external beam radiation accessible, with the majority of the installed medical physicists, and at least four to five radiation oncologists. Many units in Egypt (119 units) and South Africa (97 units). Cobalt-60 units of the centers in this area fare poorly in comparison to this standard. The and linear accelerators are both present in 50% of all countries. Only IAEA also suggests one medical physicist for every 400 new patients and Fig. 1. Radiotherapy and Brachytherapy equipment per million population in Africa, Western Europe and North America. 3 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 Table 1 Current status of radiotherapy resources in 33 African countries [31]. S/ Country Region Name RT MV Proton Ion Xray (kV) Brachy Therapy Total number of equipment per million Last N Centers Therapy Therapy Therapy Inc El of population Update 1 Algeria North Africa 16 37 0 0 12 1.12 2020 2 Angola Middle Africa 2 3 0 0 1 0.12 2023 3 Botswana Southern 1 1 0 0 1 0.85 2022 Africa 4 Burkina Faso West Africa 1 1 0 0 0 0.05 2021 5 Cameroon Middle Africa 3 2 0 0 0 0.08 2022 6 Cote D’Ivoire West Africa 1 2 0 0 0 0.08 2019 7 DR Congo East Africa 1 1 0 0 0 0.01 2021 8 Egypt North Africa 75 120 0 1 23 1.41 2022 9 Ethiopia East Africa 3 3 0 0 1 0.03 2022 10 Gabon Middle Africa 1 2 0 0 0 0.90 2018 11 Ghana West Africa 3 6 0 0 3 0.29 2022 12 Kenya East Africa 10 17 0 0 5 0.41 2022 13 Libya North Africa 5 8 0 0 0 1.16 2022 14 Madagascar East Africa 2 3 0 0 1 0.14 2022 15 Mali West Africa 1 1 0 0 0 0.05 2019 16 Mauritania West Africa 1 3 0 0 1 0.86 2019 17 Mauritius East Africa 1 3 0 0 1 3.15 2019 18 Morocco North Africa 30 46 0 0 10 1.52 2022 19 Mozambique Southern 1 1 0 0 0 0.03 2020 Africa 20 Namibia Southern 2 2 0 0 1 1.18 2022 Africa 21 Niger West Africa 1 1 0 0 0 0.04 2021 22 Nigeria West Africa 7 9 0 0 2 0.05 2022 23 Reunion East Africa 1 5 0 0 0 5.58 2018 24 Rwanda East Africa 1 2 0 0 0 0.15 2021 25 Senegal West Africa 2 2 0 0 1 0.18 2022 26 South Africa Southern 62 103 0 8 23 2.26 2022 Africa 27 Sudan North Africa 4 6 0 0 0 0.14 2022 28 Tanzania East Africa 4 8 0 0 4 0.20 2022 29 Togo West Africa 1 1 0 0 0 0.12 2022 30 Tunisia North Africa 14 26 0 1 4 2.62 2022 31 Uganda East Africa 1 3 0 0 1 0.09 2022 32 Zambia Southern 1 3 0 0 2 0.27 2019 Africa 33 Zimbabwe Southern 3 1 0 0 2 0.20 2022 Africa one radiation oncologist for every 200–250 new patients treated annu- for basic chatbots or as much as millions of dollars. The collecting and ally [4]. preparation of data, the use of hardware and computational resources, The Lancet Oncology issued a Report in 2015 on improving access to and the upkeep and improvement of systems all have costs connected radiation worldwide. The report indicated that many low- and middle- with their use or adoption of the technology [37]. This high cost makes it income countries (LMICs) lacked adequate radiotherapy coverage and challenging for most African countries to either expand existing or predicted that by 2035, there would be a need for 2,600 radiotherapy create new facilities. The small number of radiotherapy facilities that departments, 5,200 machines, and 55,800 radiation oncologists, medi- patients overwhelm reduces the efficiency and effectiveness of quality cal physicists, and radiotherapy technologists to meet the demand [35]. assurance programs implementation [34,38]. The implementation of With less than one external beam of radiotherapy equipment per million quality assurance in radiation treatment is particularly difficult in low- people across the continent, Africa has the least developed radiotherapy and middle-income countries because of a lack of staff training, a lack of capacity when compared to North America, which has roughly 15 ma- national norms, a lack of quality assurance tools, and a high patient chines per million people [28]. In Africa, this discrepancy is particularly throughput each day [39]. obvious, with only 25% of the need being addressed and 29 of the 54 It has been established that Africa’s radiation staff are in need of countries lacking a functioning radiation facility [36]. The discrepancy more and better training programs. Sub-Saharan African nations with is also stark in sub-Saharan Africa (excluding southern Africa), where well-established training programs include Tanzania, Zimbabwe, there are ten times fewer radiotherapy machines per million people than Ghana, South Africa, Egypt, Morocco, and Zambia, with the majority of in North America. It is predicted that 700 more radiotherapy machines nations reliant on outside training. The intended results of increased will be required to bring Africa’s capacity up to par. The incidence of human resource self-sufficiency have not been achieved by sending cancer is expected to double in Africa’s less-resourced regions by 2040, trainees to other continents since trained professionals are less likely to which will only increase the need for these services [32]. return to their place of origin after training and are more likely to stay Finance is one of the key issues limiting the development of radiation where they have been taught. This is probably due to a lack of efficient services in Africa. A sizable portion of the cost of administering radio- staff retention programs and inadequately equipped Centers that do not therapy is made up of supplies like teletherapy machines, brachytherapy correspond to the knowledge that staff trained in well-equipped facilities after loaders, and the necessary ancillary items like planning software should apply [40]. Insufficient clinical and/or radiation oncology and radioactive isotopes, as well as labor costs like those of oncologists, training programs are a significant barrier to addressing the shortage of medical physicists, radiation therapists, and nurses [11]. In wealthy radiotherapy staff in Africa [28]. countries, it is impossible to estimate the price of putting together an AI/ The applications of data science are still relatively underdeveloped in ML-based healthcare solution. However, it might cost as little as $6,000 Africa given the fact that the development of AI models for radiotherapy 4 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 Table 2 Summary Chart of major challenges radiotherapy faces in 33 African countries. Countries Less than one piece No AI Lack of human Countries with Insufficient data for of RT equipment radiotherapy resources with inadequate/ training AI models/ per million education sufficient training unreliable funding lack of AI technology population curriculum or expertise sources and resources Algeria + + + Angola + + + Botswana + + + + Burkina Faso + + + + Cameroon + + + + + Cote D’Ivoire + + + + D. R. Congo + + + + Egypt + + + + Ethiopia + + + + Gabon + + + + Ghana + + + + + Kenya + + + + Libya + + + + Madagascar + + + + Mali + + + + Mauritania + + + + Mauritius + + + + Morocco + + + Mozambique + + + + Namibia + + + Niger + + + + Nigeria + + + + + Reunion + + + (France) Rwanda + + + + + Senegal + + + + + South Africa + + + + Sudan + + + + + Tanzania + + + + + Togo + + + + Tunisia + + + Uganda + + + + + Zambia + + + + Zimbabwe + + + + practice requires a large volume of specific trainable data sets [37]. This throughout the entire workflow of radiotherapy, beginning with the is especially true for datasets with labels, which must be annotated by selection of the best radiation method, such as choosing between proton physicians or other medical professionals and are therefore expensive and photon radiation to treatment planning, plan evaluation and quality and time-consuming to gather. Due to the low adoption of electronic assurance, dose delivery, and patient care [1,42]. For example, a Clinical medical records and digitization in Africa, there are few locally gener- Decision Support System (CDSS) based on deep learning technology was ated useful data that are crucial for creating AI systems [41]. In addition created by Liang et al. [12] to provide cancer therapy alternatives by to data volume and quality, Africa also lacks professionals in AI tech- extracting and analyzing a significant amount of clinical data from nology. A lack of infrastructure and resources for education may prevent medical records. The study showed that by learning from clinical big some African countries from importing AI professionals or teaching AI to data of cancer patients, deep learning can assist doctors in selecting the their own populace. Thus, it’s possible that only a select group of optimal treatment option and eventually enhance cancer patient treat- wealthy people will be able to access this knowledge, worsening ment plans [12]. inequality. Radiation treatment planning begins with precise segmentation of Many nations are currently working to create a governance policy or organs at risk (OARs) and target volumes. Deep learning convolutional legislative framework to support the use of AI in many industries. There neural network (CNN)-based auto-segmentation models have recently are no regulations governing who is responsible for unfavorable results been demonstrated to increase this process’ consistency and effective- that may arise from the use of artificial intelligence in healthcare, which ness. Based on characteristics of the position and intensity of the voxel is highly probable given how and where AI may be implemented. and neighboring voxels, these models typically categorize every voxel in However, some situations and places are not foreseen or covered by the an image as belonging to an OAR or target [6]. Similarly, Machine current legislation, making such resolutions less likely to involve the learning CNNs have been used to perform automatic segmentation of application of the law. In many African nations, this will have legal radiation targets and OARs [42]. AI technology has been exceptional in repercussions for users and patients [41]. the segmentation of different structures during the treatment planning stage [12,43–47]. 4. Overview of AI applications in radiotherapy During the treatment planning stage, a 3D-dose distribution of a plan is usually generated utilizing a variety of techniques (e.g., convolution- Artificial intelligence has gained recognition in radiotherapy prac- superposition algorithm and Monte Carlo simulation) which can be tice because most of the processes in the radiotherapy chain involve tedious and time-consuming [1]. To expedite optimization or establish computer medical image processing, calculation, and hardware control. the best achievable dose distribution from the patient image, DL can With the help of big data processing and high-performance computing, forecast the dose distribution from radiation therapy treatment [42]. Al technology can be used to automate and perform numerous tasks The distribution of the desired target dose conformity and adequate 5 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 sparing of critical structures from the treatment plan is mostly achieved of information gathered during these evaluations has made it possible to with the use of knowledge-based dose volume histograms (DVHs) [14]. create AI algorithms that can predict trends and errors, like multi-leaf Many human factors, including the selection of radiation beam angles collimator positioning errors and trends in beam symmetry, as well as and the plan’s optimization parameters, affect the quality of the dose automatically identify imaging artifacts. These techniques might make distribution plans [7]. The dose distribution index (DDI) variable, which the QA process more effective, giving medical physicists more time for offers dosimetric estimates on the target coverage, organs at risk, and other tasks [7]. Commissioning and quality assurance (QA) for linear healthy tissue in the treated organ, is used to assess the quality of a accelerators (linacs) need a lot of work and time. Machine learning can treatment plan. With the aid of AI, the DDI value is predicted using be used to lessen the workload associated with linac commissioning. A machine learning from the DVHs [1]. Data-based regression analysis is machine learning system can be trained using previously obtained beam used as the dose-volume histogram (DVH) estimate algorithm to create a data to simulate the inherent correlation of beam data under various knowledge-based planning (KBP) model from a previous, clinically configurations. The trained model is thus able to produce accurate and approved treatment plan [48]. The application of machine learning to dependable beam data for linac commissioning for routine radiotherapy knowledge-based DVH techniques has shown promising outcomes in the [56]. treatment planning of head and neck, pancreas, and prostate cancers Patient-specific QA activities include dosimetric measurements of [49–52]. treatment plans, in vivo dosimetry, and monitor units [14]. For Prior to treatment, target delineations, which call for image regis- extremely complex treatment plans, a dosimeter-containing phantom is tration, are performed using multimodal and four-dimensional imaging used to physically measure the delivered dose, which is then compared modalities. Image registration is the process of identifying geometric to the intended dose. The vast majority of plans succeed in this QA stage, correspondences between two or more imaging data sets that differ in but in the rare instance where a plan is unsuccessful, numerous potential time, place, modality or subject [53]. Deep learning AI algorithms have contributing causes necessitate examination, which could postpone the ability to immediately discover the best attributes for registration treatment. In order to forecast QA passing rates based on the treatment from the input data. Using a twelve-element output vector modified plan itself and to pinpoint probable mistake sources, an AI system has from 2-dimensional (2D) DenseNet, the affine image registration deep been developed, which may eventually replace the necessity for physical learning model was successfully used to predict transformation param- dose measurements [10]. eters and register three-dimesional (3D) images. An encoder was initially applied to the image pair to extract features. Many fully con- 5. How ready is Africa for the introduction of AI in nected layers were then concatenated using the attributes as input, and radiotherapy? regression was utilized to complete the registration process [54]. In a similar vein, CNNs are regarded as the most effective and potent deep- It may not be an easy task to implement and integrate AI technology learning techniques for medical image registration. Researchers have into clinical practice across radiation centers in Africa. This is because, had modest success in registering chest computed tomography (CT) prior to integrating the technology into clinical practice, challenges of AI images, brain CT and MR images, 2D X-ray images, and 3D CT images technology must be overcome and recommendations for the integration using CNN deep learning algorithms [55]. must be met. The process will necessitate an initial significant time and AI can create effective motion management models that take motion massive resource investment, as well as efforts to comprehend the variability into consideration, including magnitude, amplitude, fre- benefits and constraints of the technology [7]. quency, and other factors. These models can forecast respiratory Regular QA is strongly recommended after the successful imple- movements using information obtained from outside surrogate makers mentation of any AI-based program [6]. The adoption of AI in clinical [56]. If motion management techniques are not used, patient or internal practice necessitates the establishment of a Quality Management Pro- organ motion during RT treatment may increase the dose administered gram (QMP) to ensure that the QA model hasn’t been accidentally to non-target tissues. AI can be utilized to create dynamic motion changed and that it is still valid after a (small) software upgrade. For this management models that are customized to a patient, which can use, a reference data collection that reflects clinical practice should be enhance tumor tracking and halt radiation delivery to insufficient target chosen at the time of commissioning. The reference dataset should be places. To precisely measure and anticipate the tumor position in forecasted again on a frequent basis and compared to the initial pre- advance, AI algorithms could automatically adapt to intricate breathing dictions during commissioning (end-to-end performance) to ensure patterns in real-time [57]. Motion during treatment planning and dose model consistency and identify changes in the workflow [6]. delivery is most notable either in tumor expansion or contraction as well Large datasets are mostly required to train, validate, and test many as anatomical variations that might affect the doses delivered to the AI learning models, especially when deep learning (DL) algorithms are tumor and other organs. In such cases, it is necessary to re-plan treat- employed. Depending on the task (i.e., training, validation, and testing) ment (also referred to as adaptive treatment) in light of the most recent that must be performed, AI models are either developed in-house, in images of the patient’s anatomy. AI may offer tools to forecast which partnership with a vendor, or already commercially accessible [6,14]. patients need treatment adaption and the appropriate interval during Using deep learning algorithms, like CNN, requires hundreds of millions which it should take place [7]. For instance, when a patient receives of trainable data [57]. For these algorithms to be used safely, high- radiation therapy, ML can spot major alterations in their anatomy and quality data are required for training [58]. The data needs to receive foretell which of their cases would benefit from adaptive radiotherapy expert validation and broad acceptance from the computer science and [42]. medical radiation science communities [59]. It is advisable to use locally In every radiotherapy department, the medical physicist is mainly acquired input data in order that the department’s clinical guidelines responsible for performing quality assurance tasks. Quality assurance and imaging methods are preserved [6]. This can be achieved by (QA) is used to check and keep an eye on the tools and processes used in establishing an open-access national data bank to accelerate the devel- diagnosis and treatment, as well as the clinical support systems. AI can opment of AI models [60]. The data may need to be triaged after it has be used to carry out automated quality checks (QCs) that, if done been evaluated to ensure that it is a curated representation of the patient manually, would not be viable on a regular basis due to the time population and clinical practice under consideration [6]. However, commitment [42]. There are two major types of QA in radiotherapy these datasets are either frequently unavailable, ridiculously expensive, namely; machine and patient-specific QA [14]. Machine quality assur- or legally restricted [55]. It is not recommended to generalize data from ance involves evaluating the capabilities of various radiation medical developed countries to underdeveloped ones without anticipating dif- devices (e.g., linear accelerators, electronic portal imaging systems, ferences. Therefore, the best course of action appears to be to ensure onboard imaging, and computed tomography CT) [10]. The abundance equity in data representation while taking into account the geographic 6 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 variances in diseases, demographics, and health services [57]. Ideal curriculum in oncology for medical students [67]. With regard to who “owns” the data and who has the right to use it, particularly where it has a commercial value, the requirement of huge 6. Role of international institutions/bodies amounts of data for the deployment of AI technology in radiotherapy practice may pose legal and ethical questions. Although obtaining pa- The Federation of African Medical Physics Organizations (FAMPO), tient agreement for data use would be ideal, it may not be possible given established in 2009, is the regional federation of the International Or- the enormous number of patients in large datasets, especially in retro- ganization for Medical Physics (IOMP) in Africa. The Federation pro- spective contexts [61]. The application of existing laws may be one of motes growth and development of medical physics in Africa. As of 2016, the most likely solutions to overcoming legal issues [41]. The laws Tabakov (2016) had reported a total number of 700 medical physicists should be able to address issues of ownership and rights of the use of in Africa [68]. Ige et al. [27] in 2020 estimated the number of medical data. If the use of existing laws cannot address the legal issues, new laws physicists in Africa to be 1,041, nearly 50% increase in the 2016 figure. can be unanimously developed by the various government in the African Most clinical medical physicists in Africa are radiotherapy-based, with continent to address legal problems that may occur due to a breach of large inadequacy of personnel in medical imaging [69]. Since its security and privacy [62]. In the meantime, anonymizing patient data establishment, FAMPO has been actively promoting the development of must be mandatory to protect patient privacy while attempting to personnel and practice of medical physics in the region [70,71]. The first maximize the utility of the data. FAMPO Conference, organized by the Association of Medical Physicists To the best of our knowledge, no AI models based on public data- in Morocco, and held from 10 to 12 November 2023 in Marrakech, bases with radiological images have been developed using deep learning Morocco, brought together over 300 participants. in Africa [63]. Even if such models exist, they might not be totally In addition to FAMPO’s contribution, agencies like International validated. A preliminary search in five electronic databases to identify Atomic Energy Agency (IAEA), International Centre for Theoretical AI tools created between January 2000 and January 2022 utilizing pa- Physics (ICTP), International Organization for Medical Physics (IOMP) tient cohorts in Africa revealed 12 prediction tools for various cancers have played very important roles towards development of the profession and uses. Unfortunately, they have not been validated for predicting and its practice in the region. cancer outcomes in about 90% of African nations and their subsequent The IAEA provides important programs for about 176 nations, implementation in radiotherapy practice [64]. The absence of AI models including low-and-middle-income countries, to help them use nuclear is due to the lack of sufficient radiological images (or data) needed to technology peacefully. There are 45 IAEA member states that take part successfully train the deep learning algorithm [63]. In the meantime, the in these programs in the Africa region. Through its programmes and challenges with limited data could be overcome with the use of a data projects, the IAEA has supported highly the development of radiation augmentation approach, which has the potential to increase the amount medicine services, capacity building, education and training, technical of usable data by adding affine image transformations to the original assistance, etc. The education and training activities cover a variety of image sets during the data training for auto-segmentation [14]. nuclear-related subjects through in-person training sessions, workshops, The integration of AI with radiation practice has frequently been led by medical physicists. AI can support knowledge-based treatment planning in the radiotherapy process with minimum input from the Table 3 IAEA coverage in Africa [72–75]. medical physicist [14]. It is advised that AI content be included in the curriculum for training and educating aspiring medical physicists at all Project Project Title Project Objective academic institutions providing such training across Africa [63]. In Code Europe for instance, the core curriculum for training medical physics RAF6051 Strengthening Education and To strengthen and sustain nuclear students both at undergraduate and graduate levels in radiotherapy has Human Resources medicine capabilities in Africa Development for Expansion through academic education been developed to include AI. The curriculum which has been imple- and Sustainability of Nuclear programmes mented at all institutions’ training medical physicists has been endorsed Medicine Services in Africa by thirty-two European Medical Physics Societies [65]. Therefore, the RAF6054 Strengthening and Improving To improve good operating curriculum for the education and training of radiotherapy personnel in Radiopharmacy Services standards and pharmaceutical Africa must be revised to meet the recommendations of international regulation of hospital preparation of radiopharmaceuticals in order to standards. This might fill the widening teaching gap throughout the expand the range of safe and continent, personalize and sustain learning, and provide individuals effective radiopharmaceuticals with a way to upskill for the changing digital future. available in African Member States A study conducted by Ige et. al (2020) revealed that the curriculum and improve patient safety in nuclear medicine practice for formal education and training programs in radiotherapy is not RAF6055 Improving the Quality of To enhance the quality of the standardized across Africa. Also, there are not many higher education Radiotherapy in the Treatment delivery of radiotherapy services in options in fields relevant to radiotherapy. The available radiotherapy- of Frequently Occurring AFRA MS through harmonized related programs are mainly academic, with little clinical content Cancers clinical training schemes and [27]. The absence of AI radiotherapy courses in the current curriculums sensitization of policy makers RAF6056 Supporting Human Resources To strengthen the treatment of may be due to a limited number of no qualified professionals capable of Development in Radiation cancer through the training and teaching and supervising AI-related courses [4]. Since the majority of Medicine education of radiation medicine the staff in a radiotherapy department are radiation oncologists, medical professionals in AFRA States physicists, and radiation therapists [4], a crucial first step in preparing RAF6057 Strengthening the Quality of To enhance the quality of the Nuclear Medicine Services delivery of nuclear medicine in radiation therapists, medical physicists, and radiation oncologists to use AFRA States through a well- therapeutic technology in a competent and safe manner may be to established quality management incorporate AI and ML principles into radiation therapy education [66]. system. The education and training of radiotherapy personnel would however RAF6058 Strengthening the Capacities To strengthen and sustain imaging require a collaboration of African institutions and radiotherapy de- for Radiopharmacy and services in Africa through academic Medical Physics and Radiology education programmes partments to collaborate with relevant international bodies and stake- for Expansion and (radiopharmacy) and training as holders. It is necessary to make a deliberate effort to enhance cancer Sustainability of Medical well as effective diagnostic and education in medical schools. There are helpful guides regarding Imaging Services interventional radiological curricular content, such as the International Union Against Cancer’s practices 7 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 online learning, fellowship programs, and schools [24,72]. Some pro- In particular, for new radiotherapy practitioners who are just starting jects on radiation medicine being run by IAEA in collaboration with its out and may not have the full radiological education and expertise in Member States in Africa are presented in Table 3. radiotherapy operations, the application of AI based on machine The International Center for Theoretical Physics (ICTP) has also learning algorithms for image analysis could be helpful in their day-to- contributed significantly towards the education and training of clinical day work [44,77]. medical physicists at the master’s degree level. The program is co- The adoption of AI technology in clinical practice will encourage sponsored by IAEA and supported by recognized international organi- radiotherapy professionals to place a priority on continuing their edu- zations. The organizations include the International Organization for cation in the field of machine and deep learning while potentially Medical Physics (IOMP), the European Federation of Organizations in improving their abilities in manual segmentation and organ contouring. Medical Physics (EFOMP), and the Italian Association of Medical Physics Knowledge and applications of AI models in radiotherapy practice (AIFM) [72]. In the last 3 decades, ICTP’s College on Medical Physics would significantly reduce staff workload as the majority of the staff’s and training courses have successfully trained over 1,000 medical work will also be focused on macro-processes, such as assessing the physicists in about 100 low-and-middle-income countries [76]. equipment performance quality [56]. Medical physicists are those who As a result, all radiotherapy departments, educational institutions are mainly responsible for overseeing the status of equipment perfor- that train students towards providing radiotherapy services, and rele- mance in Africa through QA checks. For example, for patient alignment vant stakeholders in Africa must take a strong interest and make an and motion management, medical physicists must make sure that the earnest effort in collaborating with IAEA, international educational in- performance quality of radiotherapy machines and equipment is of high stitutions and organizations, manufacturers of radiotherapy equipment, precision and within tolerance levels [37]. Both machine learning and and developers of treatment planning software to achieve this goal. deep learning techniques will assist medical physicists in better identi- Under its new Harnessing Data Science for Health Discovery and fying QA measurement problems and developing proactive QA strate- Innovation in Africa program, the National Institutes of Health in the gies [13]. Also, the notion that radiotherapy services are not widely United States has committed approximately $74.5 million over five accessible in Africa is not entirely due to the high base cost of the years to advancing data science, accelerating innovation, and promoting equipment but rather the large load of maintenance costs when a health discoveries throughout Africa. Given these resources and in- radiotherapy machine breaks down. The likelihood that a piece of vestments, AI/ML applications’ influence on healthcare in Sub-Sahara equipment will continue to operate is frequently affected by recurring African (SSA) is possible [37,60]. To speed up the integration of AI high service and maintenance costs [34]. Annual recurrent costs, which technology in radiotherapy practice, African countries must invest include maintenance and source replacement, can be in the range of 5% heavily in skills, infrastructure (i.e., data centers), and technologies such and 15% of the initial capital investment [10]. Through quality assur- as electronic health records and cloud storage to meet up with the data ance, AI technology when implemented would save maintenance costs required for training DL and ML algorithms. Cloud-based technology can due to the ability of the technology to identify and resolve inefficiencies. be used as a tool to speed up peer review by offering a platform for the African oncologists were considering using AI-based models to pre- private distribution of planning data sets between radiation centers for dict outcomes for patients with breast, cervical, and colorectal carci- external assessment and input. The utilization of such platforms for nomas [63]. This roughly corresponds to the prevalence of malignant training and education as well as reporting, learning and assessment in neoplasms across the continent, with the five most prevalent subtypes radiotherapy practices can also make it easier for people to take part in being breast, cervical, prostate, liver, and colorectal cancers. However, international clinical studies, such as those that are managed by the no actionable AI patient-based platforms have been suggested to make it International Atomic Energy Agency [77–79]. easier to screen for, identify, and treat prostatic and hepatocellular During the peak of the COVID-19 pandemic, several international cancer in the area. So, research aiming to suggest new platforms and/or collaborations between countries, institutions and individuals were external validation of current AI-based platforms, especially for liver established to develop remedial actions and mechanisms in the fight and prostate cancer as well as other, hitherto unconsidered malig- against the disease [80–83]. The period saw a boom in research output nancies, will be beneficial for Africa [63]. towards AI applications for diagnosis and treatment of cancer cases [84–86]. 8. The role of medical physicists in radiotherapy services and AI applications 7. How AI can improve radiotherapy services in Africa In the history of radiotherapy service delivery, Medical Physicists The structural issues such as the shortage of staffing in most radio- have always played a major role when it comes to developing CT-based therapy facilities in Africa that lead to inconsistent patient results would dose calculation, treatment planning, image-guided radiation therapy, be overcome when AI technology is implemented. The workflow of ra- quality assurance, and radiation protection [41]. Notwithstanding, diation treatment is time-consuming due to the numerous manual inputs Medical Physicists have been at the forefront of the application of AI involving the medical physicist, radiation oncologist, dosimetrist, and technology to medicine, including the creation and improvement of radiation therapist [13]. Hence, when implemented, the technology will imaging technologies as well as many other innovations that have augment the limited staff across Africa by reducing their workloads and boosted the quality of radiotherapy service delivery [87]. As an allowing them to spend more time with patients and providing actual example, the knowledge-based treatment planning system, which uses patient care. machine learning algorithms to construct high-quality, automated One of the most important stages that can take a lot of time during radiotherapy plans using patient images, contours, and clinical data, was treatment planning is the contouring of malignancies and the healthy created by medical physicists [41]. When the auto-planning systems are organs surrounding them. There are no specific guidelines for contour- finished, their results must be tested before being approved in full. ing medical images due to the variety of anatomical organ characteris- However, due to each patient’s individual anatomy, the proposed plan is tics [77]. To satisfy the demand for treatment in clinics with limited usually customized and adjusted by clinical medical physicists during resources and staffing shortages, AI could be used to automate image this time. A clinically acceptable solution is reached when potential analysis to help radiation therapists, oncologists, and medical physicists problems with a particular strategy led by the medical physicist are expedite the contouring of several image portions in a single session discussed with other team members, such as oncologists, therapists, and within a fraction of the time. Additionally, AI might be able to assist the dosimetrists [41]. radiation team in navigating the challenges posed by the physiological It is the responsibility of the Medical Physicist to carry out an diversity of anatomical structures during image interpretation [44,77]. appropriate routine Quality Assurance (QA) test program with clearly 8 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 defined frequency, metrics, tolerance levels, and actions to be taken in AI machine and deep learning techniques must be introduced at both the case of test failure in order to guarantee that clinically used AI algo- undergraduate and postgraduate levels, respectively. The absence of AI rithms continue to perform with the desired level of accuracy [41]. content in the African educational curriculum is due to the limited Medical Physicists design QA tools that guarantee the finest image number of AI teaching professionals. The fourth challenge has to do with quality. The Medical Physicist performs QA utilizing properly prepared funding, which requires massive investment in AI technology resources. in-phantom film/ion chamber measurements and comparing them As African countries prepare for the introduction of AI technology into against previously published dose calculation procedures in order to radiotherapy service delivery, these challenges must be addressed validate the dose predicted by deep learning AI algorithms. They also unanimously. This means that the management of all national radio- help determine the cause of a machine failure and implement corrective therapy centers, radiotherapy training, and educational centers, gov- measures, such as calibrations or quality control tests, when AI tech- ernments, and relevant stakeholders must work together to provide the niques forecast a LINAC machine failure [13]. As part of QA activities, needed solutions toward the successful integration and implementation medical physicists are responsible for ensuring patient safety during of AI technology into radiotherapy practice. radiotherapy. Machine learning has the ability to lower the risk of ra- diation exposure to patients and the radiotherapy team without Author contributions compromising image quality during imaging procedures [40]. Medical Physicists work to ensure that they establish clinical evidence for all All authors have contributed equally to the paper. All authors have existing and new AI applications in radiotherapy service delivery. They read and agreed to the published version of the manuscript. also make sure that all operators are trained in the usage of the best imaging techniques and ensure that radiation protection measures are in Funding place [13]. Not only do routine QA activities show the relevance of Clinical This research received no external funding. Physicists, but also non-routine actions. Medical Physicists’ non-routine activities in a typical radiotherapy department include implementing novel procedures into clinical practice, commissioning treatment Declaration of Competing Interest equipment, and offering patient-specific consultations [88]. During the commissioning of treatment equipment (e.g., LINAC) for clinical use, The authors declare that they have no known competing financial Medical Physicists perform thorough measurements of dosimetric pa- interests or personal relationships that could have appeared to influence rameters required to validate the treatment planning systems in order to the work reported in this paper. arrive at the best radiation modality and treatment approach for each patient. In addition, they also create operational procedures, entering of beam data into the treatment planning system, and testing of the accu- Acknowledgments racy of that system [89]. In the same way, careful commissioning into radiotherapy clinical The authors would like to acknowledge the global medical physics practice is necessary for any AI technology [88]. The typical method for community, professional and partner organizations for their direct and accomplishing this is to first interface the AI tool with the commercial indirect contribution for the development of the profession and treatment planning system (TPS), after which the vital patient anatomy continuous support in introducing new imaging and treatment related data are supplied to a selected workstation. However, it’s not as easy to technology and procedures in the developing regions, and particularly in commission a new radiation therapy technology as it is to assess LINAC Africa. dosimetric performance. Clinical physicists must engage and discuss the procedure with every team member within the department. When the References supply of all the patient anatomy data to the workstation is completed, AI optimal fluence maps are also imported into the TPS for the purpose [1] Chow JCL. 11. In: Lidströmer N, Ashrafian H, editors. Artificial Intelligence in Medicine. Cham: Springer International Publishing; 2020. p. 1–13. of leaf sequencing, dosage estimation, and possible fine-tuning. The [2] Lombe D, M’ule BC, Msadabwe SC, Chanda E. Gynecological radiation oncology in treatment team led by the clinical medical physicist works together to sub-Saharan Africa: status, problems and considerations for the future. evaluate and validate the workflow and performance of the technology International Journal of Gynecologic. Cancer 2022;32(3):451–6. [88,90,91]. As a result, Medical Physicists would continue to play a [3] International Atomic Energy Agency, Selecting Megavoltage Treatment Technologies in External Beam Radiotherapy, IAEA Human Health Reports No. 17, leading role in the optimization of radiotherapy procedures when AI IAEA, Vienna (2022). technology is eventually introduced and implemented across all radio- [4] Rosenblatt E, Zubizarreta E, editors. Radiotherapy in cancer care: facing the global therapy centers in Africa. challenge. Vienna: International Atomic Energy Agency; 2017. [5] Nwagwu WE, Adegunwa GO, Soyannwo OA. ICT and collaborative management of terminal cancer patients at the University College Hospital, Ibadan. Nigeria Health 9. Conclusions Technol 2013;3:309–25. https://doi.org/10.1007/s12553-013-0063-6. [6] Vandewinckele L, Claessens M, Dinkla A, Brouwer C, Crijns W, Verellen D, et al. Overview of artificial intelligence-based applications in radiotherapy: The purpose of this study was to assess the readiness of Africa for the Recommendations for implementation and quality assurance. Radiother Oncol introduction of Artificial Intelligence Technology into radiotherapy 2020 Dec;1(153):55–66. service delivery. To achieve this aim, a narrative review was conducted [7] Huynh E, Hosny A, Guthier C, Bitterman DS, Petit SF, Haas-Kogan DA, et al. Artificial intelligence in radiation oncology. Nat Rev Clin Oncol 2020 Dec;17(12): from relevant databases to identify challenges that might limit the 771–81. introduction of AI technology into radiotherapy practice. The first [8] GLOBOCAN 2020: New Global Cancer Data. Cited on 20th February, 2023. Last challenge identified was insufficient data needed to adequately train and update: Monday 15 February 2021. Data base: https://www.uicc.org/news/ globocan-2020-new-global-cancer-data. validate AI models. The lack of sufficient data is a result of the un- [9] Vanherk M. Errors and margins in radiotherapy. Semin Radiat Oncol 2004;14(1): availability of national data banks in all African countries as well as 52–64. resources of AI technology such as storage clouds. The second challenge [10] Simon L, Robert C, Meyer P. Artificial intelligence for quality assurance in radiotherapy. Cancer/Radiothérapie 2021 Oct 1;25(6–7):623–6. was legal and ethical issues related to rights and ownership of the data. [11] Balogun O, Rodin D, Ngwa W, Grover S, Longo J. Challenges and Prospects for The third major issue identified was the exclusion of AI content in the Providing Radiation Oncology Services in Africa. Semin Radiat Oncol 2017;27(2): educational curriculum for training radiotherapy personnel/students. 184–8. [12] Liang G, Fan W, Luo H, Zhu X. The emerging roles of artificial intelligence in To make sure students are aware of the clinical applications, potential cancer drug development and precision therapy. Biomed Pharmacother 2020 Aug; advantages, and related hazards, fundamental and advanced courses in 1(128):110255. 9 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 [13] Tran KA, Kondrashova O, Bradley A, Williams ED, Pearson JV, Waddell N. Deep [44] Lustberg T, van Soest J, Gooding M, Peressutti D, Aljabar P, van der Stoep J, et al. learning in cancer diagnosis, prognosis and treatment selection. Genome Med 2021 Clinical evaluation of atlas and deep learning based automatic contouring for lung Dec;13(1):1–7. cancer. Radiother Oncol 2018 Feb 1;126(2):312–7. [14] Manson EN, Mumuni AN, Fiagbedzi EW, Shirazu I, Sulemana H. A narrative review [45] Jackson P, Hardcastle N, Dawe N, Kron T, Hofman MS, Hicks RJ. Deep learning on radiotherapy practice in the era of artificial intelligence: how relevant is the renal segmentation for fully automated radiation dose estimation in unsealed medical physicist? source therapy. Front Oncol 2018 Jun;14(8):215. [15] Marcu LG, Marcu D. Points of view on artificial intelligence in medical [46] Hu P, Wu Fa, Peng J, Liang P, Kong D. Automatic 3D liver segmentation based on imaging—one good, one bad, one fuzzy. Health Technol 2021;11:17–22. https:// deep learning and globally optimized surface evolution. Phys Med Biol 2016;61 doi.org/10.1007/s12553-020-00515-5. (24):8676–98. [16] Elsayed Z, Lalya I, AlHussain H, Mula-Hussain L. 16. In: Al-Shamsi HO, Abu- [47] Ibragimov B, Toesca D, Chang D, Koong A, Xing L. Combining deep learning with Gheida IH, Iqbal F, Al-Awadhi A, editors. Cancer in the Arab World. Singapore: anatomical analysis for segmentation of the portal vein for liver SBRT planning. Springer Singapore; 2022. p. 445–60. Phys Med Biol 2017 Nov 10;62(23):8943. [17] International Atomic Energy Agency (IAEA). History of radiotherapy – a short [48] Ahn SH, Kim E, Kim C, Cheon W, Kim M, Lee SB, et al. Deep learning method for introduction. Copyright © 2010-2016. Accessed on 1st March, 2023. Data base: prediction of patient-specific dose distribution in breast cancer. Radiat Oncol 2021 https://humanhealth.iaea.org/HHW/MedicalPhysics/Radiotherapy/ Dec;16:1–3. Topicsofspecialinterest/HistRT/index.html. [49] Kostovski A. Artificial Intelligence in Radiotherapy. Radiološki vjesnik: radiologija, [18] Nwankwo KC, Dawotola DA, Sharma V. Radiotherapy in Nigeria: current status radioterapija, nuklearna medicina 2021;25:37–40. and future challenges. West African J Radiol 2013 Jul 1;20(2):84. [50] Xu H, Lu J, Wang J, Fan J, Hu W. Implement a knowledge-based automated dose [19] Taku N, Polo A, Zubizarreta EH, Prasad RR, Hopkins K. External beam volume histogram prediction module in Pinnacle3 treatment planning system for radiotherapy in Western Africa: 1969–2019. Clin Oncol 2021 Dec 1;33(12): plan quality assurance and guidance. J Appl Clin Med Phys 2019;20(8):134–40. e511–20. [51] Tsang DS, Tsui G, McIntosh C, Purdie T, Bauman G, Dama H, et al. A pilot study of [20] Ndonye PK, Tagoe SN. Current Status of Radiotherapy Services in Kenya. J Cancer machine-learning based automated planning for primary brain tumours. Radiat Ther 2022 Apr 6;13(4):218–33. Oncol 2022;17(1). [21] Mulape K. Perspective on radiation therapy in Zambia involving alternative [52] Jiang S, Xue Yi, Li M, Yang C, Zhang D, Wang Q, et al. Artificial Intelligence-Based technologies. Acessed on 1st March 2023. Data base: WINS 2021. https://www. Automated Treatment Planning of Postmastectomy Volumetric Modulated Arc wins.org/wp-content/uploads/2021/08/Success-story-Zambia-Final.pdf. Radiotherapy. Frontiers. Oncology 2022;12. [22] WHO Africa (2021). Rwanda Cancer Centre inaugurated on the World Cancer Day [53] Teuwen J, Gouw ZA, Sonke JJ. Artificial Intelligence for Image Registration in 2020 by President Paul Kagame. Accessed on 1st March, 2023. Data base: https:// Radiation Oncology. Semin Radiat Oncol 2022;32(4):330–42. www.afro.who.int/news/rwanda-cancer-centre-inaugurated-world-cancer-day- [54] Islam KT, Wijewickrema S, O’Leary S. A deep learning based framework for the 2020-president-paul-kagame. registration of three dimensional multi-modal medical images of the head. Sci Rep [23] Levin CV, Sitas F, and Odes RA, 1994. Radiation therapy services in South Africa. 2021;11:1860. South African Medical Journal., 1994 Sept 30;84(6):349-351. [55] Zou M, Hu J, Zhang H, Wu Xi, He J, Xu Z, et al. Rigid medical image registration [24] Bishr MK, Zaghloul MS. Radiation therapy availability in Africa and Latin America: using learning-based interest points and features. Comput Mater Continua 2019;60 two models of low and middle income countries. International Journal of Radiation (2):511–25. Oncology* Biology*. Physics 2018 Nov 1;102(3):490–8. [56] Zhao W, Shen L, Islam MT, Qin W, Zhang Z, Liang X, et al. Artificial intelligence in [25] Boyle P, Ngoma T, Sullivan R, Ndlovu N, Autier P, Stefan C, et al. The State of image-guided radiotherapy: a review of treatment target localization. Quant Oncology in Africa, 2015. Int Prev Res Institute Lyon 2016;2016:1–560. Imaging Med Surg 2021;11(12):4881–94. [26] Elmore SN, Rubio JP, Bourque JM, Pynda Y, Van der Merwe D, Zubizarreta E, et al. [57] Santoro M, Strolin S, Paolani G, Della Gala G, Bartoloni A, Giacometti C, et al. Radiotherapy Resources in Africa: An Analysis of Progress and Projected Needs for Recent applications of artificial intelligence in radiotherapy: Where we are and 2030. Int J Radiat Oncol Biol Phys 2019 Sep 1;105(1):E453. beyond. Appl Sci 2022 Mar 22;12(7):3223. [27] Ige TA, Jenkins A, Burt G, Angal-Kalinin D, McIntosh P, Coleman CN, et al. [58] Niklas L, Hutan A. Artificial Intelligence in Medicine. 1st ed. Springer; 2022 Feb Surveying the challenges to improve linear accelerator-based radiation therapy in 19. Africa: A unique collaborative platform of all 28 African countries offering such [59] Murphy A, Liszewski B. Artificial intelligence and the medical radiation profession: treatment. Clin Oncol 2021 Dec 1;33(12):e521–9. how our advocacy must inform future practice. J Med Imaging Radiat Sci 2019 Dec [28] Stecklein SR, Taniguchi CM, Melancon AD, Lombe D, Lishimpi K, Banda L, et al. 1;50(4):S15–9. Radiation sciences education in Africa: an assessment of current training practices [60] Boon IS, Lim JS, Yap MH, Yong TP, Boon CS. Artificial intelligence and soft skills in and evaluation of a high-yield course in radiation biology and radiation physics. radiation oncology: data versus wisdom. J Med Imaging Radiat Sci 2020;51(4): JCO Global Oncol 2020;(6):1631–8. S114–5. [29] Fisher BJ, Daugherty LC, Einck JP, Suneja G, Shah MM, Dad LK, et al. Radiation [61] Lee H, Kim H, Kwak J, Kim YS, Lee SW, Cho S, et al. Fluence-map generation for oncology in Africa: improving access to cancer care on the African continent. Int J prostate intensity-modulated radiotherapy planning using a deep-neural-network. Radiat Oncol Biol Phys 2014;89(3):458–61. Sci Rep 2019;9(1). [30] Elmore SN, Polo A, Bourque JM, Pynda Y, van der Merwe D, Grover S, et al. [62] Zhang D. Big data security and privacy protection. Atlantic Press; 2018. Radiotherapy resources in Africa: an International Atomic Energy Agency update [63] Andersson J, Nyholm T, Ceberg C, Almén A, Bernhardt P, Fransson A, et al. and analysis of projected needs. Lancet Oncol 2021 Sep 1;22(9):e391–9. Artificial intelligence and the medical physics profession – A Swedish perspective. [31] The IAEA Directory of Radiotherapy Centres (DIRAC), https:/dirac.iaea.org/ Phys Med 2021;88:218–25. (accessed on February 10, 2023). [64] Adeoye J, Akinshipo A, Thomson P, Su YX. Artificial intelligence-based prediction [32] Rick T, Habtamu B, Tigeneh W, Abreha A, van Norden Y, Grover S, et al. Patterns of for cancer-related outcomes in Africa: Status and potential refinements. J Global care of cancers and radiotherapy in Ethiopia. J Global Oncol 2019;5:1–8. Health 2022;12. [33] Stefan DC. Childhood cancer in Africa: an overview of resources. J Pediatr Hematol [65] European Federation of Organizations for Medical Physics. CORE CURRICULUM Oncol 2015;37(2):104–8. FOR MEDICAL PHYSICS EXPERTS ІN RADIOTHERAPY, 3rd Revised Edition. Mar [34] Vanderpuye V, Hammad N, Martei Y, Hopman WM, Fundytus A, Sullivan R, et al. 3rd, 2022. Accessed on February 18, 2023. Data Base: https://www.estro.org/ Cancer care workforce in Africa: perspectives from a global survey. Infectious ESTRO/media/ESTRO/Congresses/ICHNO/2021/Radiotherapy-CC.pdf. Agents Cancer 2019;14(1). [66] Chamunyonga C, Edwards C, Caldwell P, Rutledge P, Burbery J. The impact of [35] Barksby R. Expanding access to radiotherapy in sub-Saharan Africa. Lancet Oncol artificial intelligence and machine learning in radiation therapy: considerations for 2020 Aug 1;21(8):1019. future curriculum enhancement. J Med Imaging Radiat Sci 2020 Jun 1;51(2): [36] Efstathiou JA, Heunis M, Karumekayi T, Makufa R, Bvochora-Nsingo M, Gierga DP, 214–20. et al. Establishing and delivering quality radiation therapy in resource-constrained [67] Barton MB, Frommer M, Shafiq J. Role of radiotherapy in cancer control in low- settings: the story of Botswana. J Clin Oncol 2016;34(1):27–35. income and middle-income countries. Lancet Oncol 2006 Jul 1;7(7):584–95. [37] Waljee AK, Weinheimer-Haus EM, Abubakar A, Ngugi AK, Siwo GH, Kwakye G, [68] Tabakov S. Global Number of Medical Physicists and its Growth 1965–2015. J Med et al. Artificial intelligence and machine learning for early detection and diagnosis Phys Int 2016;4:78–81. of colorectal cancer in sub-Saharan Africa. Gut 2022;71(7):1259–65. [69] Trauernicht C, Hasford F, Khelassi-Toutaoui N, Bentouhami I, Knoll P, Tsapaki V. [38] Swanson W, Kamwa F, Samba R, Ige T, Lasebikan N, Mallum A, et al. Medical physics services in radiology and nuclear medicine in Africa: challenges Hypofractionated radiotherapy in African cancer centers. Front Oncol 2021 Feb;19 and opportunities identified through workforce and infrastructure surveys. Heal (10):618641. Technol 2022 Jul;12(4):729–37. https://doi.org/10.1007/s12553-022-00663-w. [39] Meghzifene A. Medical physics challenges for the implementation of quality [70] The first regional Conference on the Federation of African Medical Physics assurance programmes in radiation oncology. Clin Oncol 2017;29(2):116–9. Organizations. Accessed on 10th March, 2023. Published 2022. Database: http:// [40] Ndlovu N. Radiotherapy treatment in cancer control and its important role in www.acmp2022.com/. Africa. Ecancermedicalscience 2019:13. [71] Federation of African Medical Physics Organizations. Promoting the Application of [41] Owoyemi A, Owoyemi J, Osiyemi A, Boyd A. Artificial intelligence for healthcare Physics in medicine in the African region. Accessed on 10th March, 2023. in Africa. Frontiers in Digital Health 2020;2:6. Published 2023. Database: https://fampo-africa.org/. [42] Avanzo M, Trianni A, Botta F, Talamonti C, Stasi M, Iori M. Artificial intelligence [72] IAEA Technical Cooperation in Africa International Atomic Energy Agency and the medical physicist: welcome to the machine. Appl Sci 2021 Feb 13;11(4): Department of Technical Cooperation 2018. Accessed on 11th March 2023. 1691. Database: https://www.iaea.org/sites/default/files/20/01/tc-africa-2018.pdf. [43] Ibragimov B, Xing L. Segmentation of organsat-risks in head and neck CT images [73] IAEA Technical Cooperation Programme Cycle Management Framework. Accessed using convolutional neural networks. Med Phys 2017;44:547–57. on 16th March 2023. Database: https://pcmf.iaea.org/MyTCPRIDE.aspx. 10 E.N. Manson et al. P h y s i c a M e d i c a 113 (2023) 102653 [74] IAEA Technical Cooperation Programme Cycle Management Framework on radiation medicine professionals in Ghana. Heal Technol 2021;11:119–26. https:// RAF6055. Accessed on 17th March 2023. Database: Database:http://pcmf.iaea. doi.org/10.1007/s12553-020-00507-5. org/MyTCPRIDE/General/TCPRIDEProject.aspx?ProjectNumber=RAF6055. [82] Hasford F, Ige TA, Trauernicht C. Safety measures in selected radiotherapy centres [75] IAEA Technical Cooperation Programme Cycle Management Framework on within Africa in the face of Covid-19. Heal Technol 2020;10(6):1391–6. https:// RAF6056. Accessed on 17th March 2023. Database: http://pcmf.iaea.org/ doi.org/10.1007/s12553-020-00472-z. MyTCPRIDE/General/TCPRIDEProject.aspx?ProjectNumber=RAF6056. [83] Vassileva J, Holmberg O. IAEA support to the radiation protection of patients in the [76] Tabakov S, Stoeva M. Collaborative Networking and Support for Medical Physics time of the COVID-19 global pandemic. Heal Technol 2022 May;12(3):637–41. Development in Low and Middle Income (LMI) Countries. Health Technol 2021;11: https://doi.org/10.1007/s12553-022-00659-6. 963–9. https://doi.org/10.1007/s12553-021-00591-1. [84] Caroline B. AI being developed to help cancer patients during the COVID-19 [77] Lewis PJ, Amankwaa-Frempong E, Makwani H, Nsingo M, Addison ECDK, pandemic. Accessed on 19th March, 2023. Published 2020. Database: https:// Acquah GF, et al. Radiotherapy planning and peer review in sub-saharan Africa: A www.imperial.ac.uk/news/197433/ai-being-developed-help-cancer-patients/. needs assessment and feasibility study of cloud-based technology to enable remote [85] Sebastian AM, Peter D. Artificial Intelligence in Cancer Research: Trends, peer review and training. JCO Global Oncol 2021;(7):10–6. Challenges and Future Directions. Life 2022;12(12):1991. https://doi.org/ [78] Martin, C. J., Kron, T., Vassileva, J., Wood, T. J., Joyce, C., Ung, N. M., Small, W., 10.3390/life12121991. Gros, S., Roussakis, Y., Plazas, M. C., Benali, A.-H., Djukelic, M., Ragab, H., & [86] Chinnam S, Rathore S, Jain V. In: Artificial Intelligence and Computational Abuhaimed, A. (2021). An international survey of imaging practices in Dynamics for Biomedical Research. De Gruyter; 2022. p. 117–44. radiotherapy. Physica Medica: PM: An International Journal Devoted to the [87] Feng M, Valdes G, Dixit N, Solberg TD. Machine learning in radiation oncology: Applications of Physics to Medicine and Biology: Official Journal of the Italian opportunities, requirements, and needs. Front Oncol 2018 Apr;17(8):110. Association of Biomedical Physics (AIFB), 90, 53–65. 10.1016/j.ejmp.2021.09.004. [88] Apex Physics Partners. How to Prepare for AI in Medical Physics. Cited on March 2, [79] Zarei, M., Gershan, V., & Holmberg, O. (2023). Safety in radiation oncology 2023. Published in 2022. Database: https://apexphysicspartners.com/wp-content/ (SAFRON): Learning about incident causes and safety barriers in external beam uploads/2022/08/How-to-Prepare-for-AI-in-Medical-Physics.pdf. radiotherapy. Physica Medica: PM: An International Journal Devoted to the [89] Tang X, Wang B, Rong Yi. Artificial intelligence will reduce the need for clinical Applications of Physics to Medicine and Biology: Official Journal of the Italian medical physicists. J Appl Clin Med Phys 2018;19(1):6–9. Association of Biomedical Physics (AIFB), 111(102618), 102618. 10.1016/j. [90] Palta JR. Linear accelerator acceptance testing and commissioning. Med Phys 2003 ejmp.2023.102618. Jun 1;30(6):1356–7. [80] Ng KH, Stoeva MS, editors. Medical Physics During the COVID-19 Pandemic: [91] Li X, Sheng Y, Wu QJJ, Ge Y, Wang C, Brizel DM, et al. Commissioning of an Global Perspectives in Clinical Practice, Education and Research. CRC Press; 2021 Artificial Intelligence (AI) Tool for Automated Head and Neck Intensity Modulated Mar 28. 10.1201/9781003144380. Radiation Therapy (IMRT) Treatment Planning. Int J Radiat Oncol Biol Phys 2022; [81] Hasford F, Sosu EK, Awua AK, Rockson P, Hammond EN. Knowledge and 114(3):S39. perception on the transmission and control of SARS-COV-2 infection among allied 11