The Journal of Nutrition 153 (2023) 2328–2338journal homepage: https://jn.nutrition.org/Nutrient Requirements and Optimal NutritionValidation of Mobile Artificial Intelligence Technology–Assisted
Dietary Assessment Tool Against Weighed Records and 24-Hour Recall
in Adolescent Females in Ghana
Gloria K. Folson 1,*, Boateng Bannerman 1, Vicentia Atadze 1, Gabriel Ador 1, Bastien Kolt 2,
Peter McCloskey 3, Rohit Gangupantulu 3, Alejandra Arrieta 2, Bianca C. Braga 4,
Joanne Arsenault 5, Annalyse Kehs 3, Frank Doyle 3, Lan Mai Tran 6, Nga Thu Hoang 7,
David Hughes 3, Phuong Hong Nguyen 2,8, Aulo Gelli 2
1 Noguchi Memorial Institute for Medical Research, University of Ghana, Accra, Ghana; 2 International Food Policy Research Institute, Washington,
DC, United States; 3 Penn State University, State College, PA, United States; 4 Friedman School of Nutrition Policy and Science, Tufts University,
Medford, MA, United States; 5 Intake – Center for Dietary Assessment, FHI Solutions, Washington, DC, United States; 6 Emory University, Atlanta,
GA, United States; 7 National Institute of Nutrition, Hanoi, Vietnam; 8 Thai Nguyen University of Pharmacy and Medicine, Thai Nguyen, VietnamA B S T R A C T
Background: Important gaps exist in the dietary intake of adolescents in low- and middle-income countries (LMICs), partly due to expensive
assessment methods and inaccuracy in portion-size estimation. Dietary assessment tools leveraging mobile technologies exist but only a few
have been validated in LMICs.
Objective: We validated Food Recognition Assistance and Nudging Insights (FRANI), a mobile artificial intelligence (AI) dietary assessment
application in adolescent females aged 12–18 y (n ¼ 36) in Ghana, against weighed records (WR), and multipass 24-hour recalls (24HR).
Methods: Dietary intake was assessed during 3 nonconsecutive days using FRANI, WRs, and 24HRs. Equivalence of nutrient intake was
tested using mixed-effect models adjusted for repeated measures, by comparing ratios (FRANI/WR and 24HR/WR) with equivalence
margins at 10%, 15%, and 20% error bounds. Agreement between methods was assessed using the concordance correlation coefficient
(CCC).
Results: Equivalence for FRANI and WR was determined at the 10% bound for energy intake, 15% for 5 nutrients (iron, zinc, folate, niacin,
and vitamin B6), and 20% for protein, calcium, riboflavin, and thiamine intakes. Comparisons between 24HR and WR estimated equivalence
at the 20% bound for energy, carbohydrate, fiber, calcium, thiamine, and vitamin A intakes. The CCCs by nutrient between FRANI and WR
ranged between 0.30 and 0.68, which was similar for CCC between 24HR and WR (ranging between 0.38 and 0.67). Comparisons of food
consumption episodes from FRANI and WR found 31% omission and 16% intrusion errors. Omission and intrusion errors were lower when
comparing 24HR with WR (21% and 13%, respectively).
Conclusions: FRANI AI–assisted dietary assessment could accurately estimate nutrient intake in adolescent females compared with WR in
urban Ghana. FRANI estimates were at least as accurate as those provided through 24HR. Further improvements in food recognition and
portion estimation in FRANI could reduce errors and improve overall nutrient intake estimations.
Keywords: adolescent, diet, dietary assessment tools, dietary intake assessment, dietary methods, low-income countries, validationAbbreviations: AI, artificial intelligence; CCC, concordance correlation coefficient; EAR, estimated average requirement; FRANI, Food Recognition Assistance and
Nudging Insights; IOM, Institute of Medicine; LIMC, low- and middle-income countries; LOA, limits of agreement; MPA, mean probability of adequacy; WR, weighed
records; 24HR, 24-hour recall.
* Corresponding author. E-mail address: gfolson@noguchi.ug.edu.gh (G.K. Folson).
https://doi.org/10.1016/j.tjnut.2023.06.001
Received 1 February 2023; Received in revised form 31 May 2023; Accepted 2 June 2023; Available online 3 June 2023
0022-3166/© 2023 The Authors. Published by Elsevier Inc. on behalf of American Society for Nutrition. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338Introduction
Dietary risk factors are estimated to cause 11 million deaths
and 255 million disability-adjusted life-years yearly [1].
Although insufficient nutrient intakes are the key risk factors for
undernutrition [2], rapid increases in unhealthy diets, coupled
with reductions in physical activity have contributed to increases
in rates of overweight and obesity [3]. Malnutrition during
childhood and adolescence is associated with psychosocial
problems, social stigmatization, poor self-image [4], non-
communicable diseases in adulthood [5–7], and reduced life
expectancy [8,9]. Adolescence is also a time when habits are
formed when youth start making their own decisions and are
particularly sensitive and selective about information sources for
decisions. Despite these challenges, there are no global nutrition
targets for adolescents, and adolescent nutrition has been mostly
overlooked in interventions and policies [10]. Female adoles-
cents are of particular interest because in many low- and
middle-income countries, the double burden of malnutrition is
high in this group, leading to poor health outcomes for the
adolescent, which are passed on to the next generation [11].
Data on food and nutrient consumption are essential to
inform nutrition policies and programs. However, there are
important gaps in the data on diets in low- and middle-income
countries (LIMCs), particularly for adolescents [12,13]. Quanti-
tative dietary intake data collection and analysis are complex
and expensive. Survey-based dietary assessment commonly in-
volves the multipass 24-hour recall (24HR) method that has been
validated in LMICs for adults reporting their intake and that of
their young children and adolescents [14]. In children and ad-
olescents, however, the accuracy of self-reported food intake
without caregiver assistance is not clear. Several subject- and
assessment-related constraints have been identified, including
the ability and willingness to self-report food consumption, and
the higher variability in their daily nutrient intakes compared
with adults [15]. From a costing perspective, undertaking mul-
tipass 24HR requires substantial financial investments of the
order of $60–$100 per person recall [16].
To address gaps in dietary assessment methods, mobile pho-
ne–assisted dietary assessment tools have been proposed in some
studies, but existing tools are constrained by a lack of assess-
ments of validity and feasibility of use in LMIC, including in
adolescents [17]. Nudging for Good (NFG) is an innovation
aimed at developing, validating, and examining the feasibility of
using AI mobile technology to provide real-time diagnostics and
tailored "nudging" as a strategy to improve the diets and nutri-
tion of adolescent females living in urban Ghana and Vietnam
[18]. The NFG project design has 3 main stages. The first stage
had 3 components: preparing the food inventory, taking pictures
of the foods we cooked, and finally annotating these pictures. In
the second stage, semantic segmentation models were trained to
recognize foods and estimate portion sizes using the annotated
images from the first stage. Finally, the Food Recognition
Assistance and Nudging Insights (FRANI) mobile application was
developed. This involved 1) conducting 2 rounds of focus group
discussions with potential users to develop the front-end user
interface; and 2) developing an Android-based mobile phone
application integrating the user interface with the AI model and
back-end databases.2329This study aimed to validate FRANI, which is the new mobile
AI application for dietary assessment in adolescent females aged
12–18 y in Ghana against the gold standard of observed weighed
records (WR) and compare the accuracy of FRANI with the 24HR
method. Specifically, the study objectives were 1) to estimate
nutrient intake and adequacy of micronutrient intake using the 3
methods; 2) to assess the equivalence bounds and extent of
agreement of FRANI with WR and 24HR methods; and 3) to
examine the sources of error (memory, omission, intrusion, and
portion estimation) for FRANI and 24HR methods.Methods
Study population
The intervention targeted 10 urban communities that were
involved in the Accra Urban Adolescent Nutrition Study in
Ghana (NMIMR-IRB Study Number 022–19–20). The Accra
Urban Adolescent Nutrition Study employed purposive sampling
according to socioeconomic profile of neighborhoods to provide
a mix of different food environments and included a total of 10
neighborhoods: 5 neighborhoods were randomly selected from
the list of neighborhoods belonging to the lowest tertile of
housing (representing the poorest neighborhoods) using the
quality/slum index [19,20] and 5 neighborhoods selected from
the highest tertile of housing quality/slum index (representing
better-off neighborhoods). The 10 neighborhoods included
Abelenkpe, Accra New Town, Achimota, Kokomlemle, New
Mamprobi, Nungua Old Town, Chorkor, South Industrial Area,
South La Estates, and West Legon Residential (Supplemental
Figure 2). A random sample of households with at least 1
adolescent girl in the target age range was drawn using the
census lists compiled by the Accra Urban Adolescent Nutrition
Study team.Sample size
For this study, a sample of 36 adolescent girls aged 12–18 y
were purposively drawn for representation from different so-
cioeconomic classes from the Accra Urban Adolescent Nutrition
Study. For logistical reasons, 5 out of the 10 locations were
purposively selected and participants were randomly drawn as
follows: Achimota (9), Kokomlemle (5), Nungua (5), West Legon
(6), and Chorkor (6). Participants were predominantly of low
socioeconomic status. Hence a further purposive selection was
made of 5 participants from Korle-Bu, a higher socioeconomic
area.
The sample size was based on the ability to detect a 10%
difference in energy intake estimated using the different dietary
assessment methods and detecting equivalence within 10%
bounds, as shown in a validation study in a similar study popu-
lation [18]. We also aimed to assess whether the different
methods were equivalent and recalculated the bound we could
detect equivalence for given our sample size and actual mean
and SD of energy intakes by weighed record, and our sample size
was sufficient to detect a difference in energy intake within a
13% bound (alpha ¼ 0.05, beta ¼ 20%).
Validation studies in the literature have suggested using
equivalence bounds of 10% or 15% of the difference between
methods as reasonable for nutrient intake comparisons [21,22],
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338although 1 dietary intake validation study has used a bound of
20% [23] Following Arsenault et al. [14] and Nguyen et al. [24],
given uncertainties about what the acceptable bound should be,
and what could be detected with our sample size, we also
considered bounds of 15% and 20%, which have been used in
other dietary validation contexts [21]. It is important to note that
the same criteria for equivalence margins (or bounds) were used
across all the nutrients. In addition to using equivalence bounds,
we also used the concordance correlation coefficient (CCC) to
assess the extent of agreement in nutrient intake estimates across
the assessment methods [25]. This approach follows our vali-
dation study in Vietnam [24].
Data collection
Dietary intake was assessed on 3 nonconsecutive days,
including 2 weekdays and a weekend day for each of the 3 di-
etary assessment methods, while ensuring that the reference day
was the same for each method. For example, on the first weekday
visit, trained enumerators visited the participant early in the
morning to handover the mobile phone application and then
began the WR including all foods and beverages consumed. On
the following day, a different enumerator undertook a multipass
24HR with the same participant. Data collection for WR and
FRANI application thus took place simultaneously on each
reference day, whereas the 24HR was completed the subsequent
day using the previous day as a reference. A team of 14 enu-
merators and 2 supervisors who had nutrition or health-related
backgrounds and previous experience in dietary data collection
were trained over 13 d, using lectures, role play, mock in-
terviews, and field practice methods. Training to administer the
24HRmultipass questionnaire included prompting for frequently
forgotten foods. All data were checked daily by supervisors for
completeness and accuracy.
Informed consent was requested from each of the participants
using a standardized form. Consent/accent for adolescent girls
was provided by themselves, and parental consent was obtained
from their parents or legal guardians. Ethical clearance was ob-
tained from IFPRI IRB and Noguchi IRB. (Federal-wide Assur-
ance, FWA, number 00001824, NMIMR IRB CPN 078/19–20).
Food Recognition Assistance and Nudging Insights
Thirty-six standard android mobile phones (Samsung Galaxy
A21s) were preloaded with preconfigured AI-app and provided
to participants during a prespecified 7-d period. Participants
were trained and instructed to take images of the foods and
beverages consumed at every meal, or instance of food con-
sumption, using the FRANI mobile application. Users would take
a picture of the meal they were about to consume, confirm the
classification of food returned by FRANI and input the amount of
actual food consumed as a proportion of the total portion served.
When food classification returned by FRANI was not accurate or
complete, users could record the particular food item consumed
by selecting the appropriate item from a comprehensive list of
foods consumed in Accra compiled from previous studies. To
facilitate portion-size estimation, a “pop-socket” was used as a
standardized visual prop. A pop-socket is a small disk of standard
size (diameter¼ 4 cm) that participants placed next to food items
when taking the picture of the food they were about to consume.
FRANI food recognition incorporates an algorithm designed to
adjust each pixel in the respective images using the pop-socket2330reference to estimate the 2-dimensional area covered by each
food in the image and then estimate the weight in grams based
on that area.
Observed weighed records
Trained enumerators monitored participants from early
morning until after their last evening meal, recording the
weights of foods and liquids consumed chronologically. Enu-
merators verified whether any meals were consumed before
their arrival. If a meal had been eaten before the enumerator’s
arrival, an alternative quantification method was used through
recall. When a mixed dish was consumed, enumerators recorded
details of the recipe, including the weights of the individual in-
gredients in the dish, the final weight of the cooked mixed dish,
and the weight consumed by the individual. Details of recipes
were provided by the household member who cooked the meals.
If a food item involved residual waste (bones, skin, and others),
the enumerators recorded the weight of the waste. Enumerators
confirmed that participants had eaten their last meal of the day
before departing in the evening.
24-Hour recalls
Dietary data for the 24HR survey was collected using an
adaptation of the interactive, multipass quantitative 24HR recall
method [26]. The method involves interviewing participants and
listing all foods and liquids consumed individually, as well as
listing separately ingredients consumed as components of rec-
ipes. All foods and liquids were recorded regardless of quantities
consumed, including condiments. Interviews were conducted
with the use of visual aids to assist in estimating portion sizes. A
preferred method was established for each food, including direct
weighing of substitutes similar in nature to the consumed foods.
Dry powdery foods were estimated with “gari,” a dry powdery
food made by pan-roasting of grated cassava. In some cases,
household handy measures and calibrated portion-size models
were used to estimate weights of cooked grains, roots, and tu-
bers. Volumes of liquid/semi-solid foods such as soups, sauces,
and porridges were estimated with water. Some items were
estimated by price.
During data collection, all data were checked daily by the
supervisor, research assistants, and Principal Investigator, for
consistency and plausibility. Queries were followed up with re-
spondents for clarification within 48 h. Food density values and
household measure weights were used to convert quantity data
collected from volume or household measures into grams. Con-
version factors were applied to calculate the weights of the items
that were estimated by household measures, food models, and
water. Conversion factors and recipe information were collected
before the dietary assessment. Weights of items estimated by
price were obtained by buying and weighing them. All data entry
was checked for completeness and accuracy by 3 members of the
research team.
Estimating food and nutrient intakes
For the 3 different methods, quantities in grams of different
food items that participants had consumed over the past 24 h
were converted into nutrients using several food-composition
tables adapted to local Ghanaian foods. Nutrient composition
data were obtained from the West African food-composition
table [27] and the RING nutrient composition table [28],
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338which is a compilation of food-composition databases relevant to
Ghana. The computation of nutrients was done using a program
written in Stata. All micronutrients were expressed in micro-
grams and milligrams except vitamin A, which was expressed in
micrograms of retinol activity equivalents (RAE). For each
micronutrient, we calculated the probability of adequacy of
intake using the estimated average requirements (EAR) and SD
from the European Food Safety Authority (EFSA) recommenda-
tions [29]. For vitamin B-12, as the EFSA recommendations only
provide the adequate intake reference values, the adequacy ratio
was calculated as the ratio between the actual intake and the
adequate intake reference value, capped at 1 when the actual
intake was higher than the adequate intake reference value. The
mean probability of adequacy (MPA) was calculated by aver-
aging the probability of adequacy (or Adequacy Ratios for B-12)
values across the 11 micronutrients considered. The dietary data
were used to also assess intakes of different food groups and
measure dietary diversity of individuals. Each individual food
was classified into a food group based on the Minimum Dietary
Diversity for Women (MDD-W) guidelines whose food groups
include: 1) grains, white roots and tubers, and plantains; 2)
pulses; 3) nuts and seeds; 4) dairy; 5) flesh foods; 6) eggs; 7) dark
green leafy vegetables; 8) vitamin A-rich fruits and vegetables; 9)
other vegetables; and 10) other fruits [30]. This classification
was shown to provide a food group score that is an adequate
proxy of the MPA of nonpregnant and nonlactating women of
reproductive age. Oils (except vitamin A-enriched oils and red
palm oil), drinks, and condiments were not classified into food
groups. Foods whose total consumption during the day was less
than 10 g were considered condiments.
Statistical analysis
Bland–Altman plots were used to depict the individual dif-
ferences in energy intake for each method (recall minus
observed) versus the average energy intake by the 2 methods.
The limits of agreement were calculated as the mean difference
 1.96 standard deviations, and limits of agreement lines were
examined visually.TABLE 1
Characteristics of adolescents (n ¼ 36)
Characteristics (household) Mean (SD)/%
Number of people in household, n 4.2 (1.8)
Proportion of households owning assets:
Mobile phone 100.0
Bed 89.2
Electricity 89.2
TV 86.5
Refrigerator 62.2
Radio 56.8
Wall clock 35.1
Sewing machine 21.6
Computer/Tablet computer 16.2
Freezer 16.2
Video deck/DVD/VCD 10.8
Washing machine 8.1
HH head’s highest education level completed
Preschool 3.9
Primary school 15.4
Junior High 46.2
Senior High 26.9
Tertiary education 7.7
2331Nutrient intakes were tested for normality, and log-
transformed values were used for statistical testing. We pre-
sented descriptive statistics only on nutrient intakes for ease of
interpretation. Validation of FRANI against WRs and compari-
sons with 24HR was undertaken using equivalence testing of
mean differences of log-transformed nutrient intakes using 10%,
15%, and 20% equivalence bounds [14] using mixed-effect
models to account for the repeated measures. In addition, we
examined the extent of agreement between FRANI and WR, and
between 24HR and WR using the CCC also adjusted for repeated
measures [31].
Errors in portion estimation were assessed by comparing the
reported food amounts by 24HR to the observed amounts by WR
for the most commonly consumed foods. We considered foods
that had at least 60 consumption episodes across all children that
were present in both WR and 24HR, as this captured most
frequently consumed foods.Results
Complete recalls were obtained for a total of 99 person-days
out of the planned 108. This was due to challenges related to
the upload of images on FRANI due to 3G network availability,
resulting in the inability to reconcile the images to the reference
period in the WR assessment. As such, data from 9 person-days
were excluded from the analysis.Descriptive statistics
Adolescent female participants were 16.5 y old on average
(Table 1). All but 1 participant (3%) had completed primary
school, 38% had completed junior high school, and 16% had
completed senior high school. Participants lived in households of
4 people on average, with all households owning a mobile
phone, more than 85% of households had access to electricity
and owned a TV, and 62% owned a refrigerator. Mean daily
energy intakes were 1844, 1898, and 2066 kcals/d for WR,
FRANI application, and 24HR, respectively (Table 2). TheCharacteristics (adolescents)) Mean (SD)/%
Age (y) 16.5 (1.7)
Highest education level completed:
Preschool 2.7
Primary school 43.2
Junior High 37.8
Senior High 16.2
Most recent math exam score (0–100) 55.4 (20.7)
Most recent English exam score (0–100) 62.6 (16.7)
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338
TABLE 2
Nutrient intakes of adolescents over 3 d of observed weighed records, FRANI app and 24-h recall
Nutrient intake per person day WR (n ¼ 99)1 FRANI (n ¼ 99)1 24HR (n ¼ 99)1
Mean (SD) Median (IQR) Mean (SD) Median (IQR) Mean (SD) Median (IQR)
Energy, kcal 1844 (797) 1770 (1002) 1898 (938) 1680 (1304) 2066 (873) 1996 (1230)
Protein, g 49.8 (25.7) 43.7 (31.8) 56.6 (31.5) 51.2 (48) 60.3 (29.5) 55.5 (39.4)
Fat, g 83.1 (58.2) 70.2 (70.1) 97.9 (70.3) 68.5 (99) 95.7 (64.3) 74.4 (87.2)
Carbohydrate, g 187 (81.2) 181.1 (91.2) 175.1 (96.3) 147.5 (101.7) 202.4 (107.9) 173.9 (141.4)
Fiber, g 22.7 (12.6) 20.5 (14.9) 20.7 (13.8) 16.6 (17.8) 23.7 (13) 21.8 (16.7)
Calcium, mg 482.7 (254.1) 461.5 (337) 558.8 (350.7) 472.9 (602.5) 531.3 (286.9) 487 (337.1)
Folate, μg 259.6 (214.5) 176 (271.8) 279 (239.4) 197.8 (280.1) 296.2 (243.3) 182.1 (339.3)
Iron, mg 14.1 (7.3) 13.1 (9.3) 14.1 (8.3) 12.1 (9.2) 16.1 (8.1) 15.4 (11.8)
Niacin, mg 16.6 (10.3) 13.7 (13.2) 17.8 (12.5) 15.7 (14.7) 19.5 (13.1) 15.8 (14.4)
Riboflavin, mg 0.6 (0.3) 0.6 (0.4) 0.7 (0.4) 0.6 (0.5) 0.8 (0.4) 0.7 (0.8)
Thiamine, mg 1 (0.6) 0.8 (0.8) 0.9 (0.6) 0.8 (0.8) 1.1 (0.7) 0.9 (0.9)
Vitamin A (RE), μg 879.2 (712.5) 649.7 (714.3) 1177.3 (937.5) 879.6 (1267.2) 944.5 (755.3) 690.1 (903.6)
Vitamin B6, mg 1.5 (0.8) 1.3 (0.9) 1.6 (0.9) 1.4 (1.3) 1.7 (0.8) 1.5 (1.1)
Vitamin B12, μg 2.4 (2) 1.8 (2.9) 2.7 (2.4) 2.1 (3.1) 2.5 (1.8) 2.1 (2.8)
Vitamin C, mg 69.7 (46.6) 55.7 (49.5) 73.4 (49.3) 60.6 (61.7) 73.7 (44.4) 70.1 (63.6)
Zinc, mg 6.3 (3.1) 5.8 (3.7) 6.7 (3.8) 6.3 (5.7) 7.5 (3.3) 7.2 (4.9)
24HR, 24hr recall; FRANI, Food Recognition Assistance and Nudging Insights; IQR, interquartile range; WR, weighed records.
1 Number of person-days ¼ 99, equal to number of subjects (n ¼36) multiplied by number of completed recalls.probabilities of adequate micronutrient intake were low (<20%)
for calcium, iron, zinc, riboflavin, and vitamin A. The MPA was
estimated at 35%, 38%, and 41% for WR, FRANI, and 24HR
methods, respectively (Table 3), with the differences relative to
estimates of MPA prevalence from WR found to be statistically
significant for FRANI (Δ ¼ 3.5%, P ¼ 0.001) and 24HR (Δ ¼
6.8%, P< 0.001), and for differences between FRANI and WR (Δ
¼ 3.2%, P ¼ 0.003).
On average, adolescents consumed ~6 food groups daily and
more than 80% consumed at least 5 food groups per day. The
proportion of days when adolescents consumed different food
groups was similar among the 3 methods (Figure 1). WR
demonstrated that almost all adolescents (92%–100%)
consumed grains, meat, nuts and seeds, and other vitamin A-rich
fruits and vegetables daily. Dairy and eggs were consumed by
40% and 38% of participants, respectively. However, fruits were
only consumed by 25% of participants.
Bland–Altman plots with 95% of differences in intakes
expressed as a ratio for log-transformed data for energy and
nutrients as estimated by FRANI application compared with WRTABLE 3
Estimated probability of adequate intake among adolescent females in Acc
EAR1
Calcium, mg 960.0
Folate, μg 250.0
Iron, mg 7.0
Niacin, mg 1.3
Riboflavin, mg 0.8
Thiamine, mg 0.07
Vitamin A (RE), μg 490.0
Vitamin B12, μg AI¼4
Vitamin B6, mg 1.3
Vitamin C, mg 75.0
Zinc, mg 9.9
Mean probability of adequacy of micronutrients, %
24HR, 24hr recall; FRANI, Food Recognition Assistance and Nudging Insig
1 Estimated average requirement based on EFSA recommendations.
2 Number of person-days¼99, equal to number of subjects (¼36) multipl
2332are displayed in Supplemental Figure S1. The difference in log-
transformed nutrient intakes (FRANI or 24HR nutrient 
observed nutrient) is equivalent to the ratio or proportion
(FRANI or 24HR nutrient/observed nutrient).
Equivalence testing
When comparing ratios of log-transformed intakes from
FRANI with WR (Figure 2A and Supplemental Table S1), we
observed equivalence at the 10% bound for energy intakes
(mean ratio ¼ 1.00, 90% CI 0.91, 1.09); 15% bound for folate
(1.01, 90% CI 0.89, 1.14); iron (0.94, 90% CI 0.86, 1.04); zinc
(0.97, 90% CI 0.87, 1.08); niacin (0.98, 90% CI 0.87, 1.10); and
vitamin B6 (1.03, 90% CI 0.93, 1.15). Protein, calcium, ribo-
flavin, and thiamine intakes were found to be equivalent at the
20% bound. Similar comparisons between 24HR and WR found
equivalence at the 15% bound for total carbohydrate intake
(mean ratio ¼ 1.03, 90% CI 0.95, 1.13) and 20% bounds for
energy, fiber, calcium, thiamine, and vitamin A (Figure 2B and
Supplemental Table S1). Wide bounds (>30%) were found for
vitamin A (FRANI) and vitamin B12 (FRANI, 24HR), andra, Ghana (n ¼ 36) by WR, FRANI application, and 24HR methods
WR (n ¼ 99)2 FRANI (n ¼ 99)2 24HR (n ¼ 99)2
0.0 (0) 2.9 (1.5) 0.3 (0.3)
45.5 (4.5) 51.4 (4.9) 64.7 (10.7)
13.1 (1.6) 9.9 (1.8) 18.9 (4.0)
85.7 (2.2) 87.3 (2.5) 91.2 (4.6)
0.1 (0.1) 0.3 (0.4) 0 .0(0.1)
91 (2.4) 93.1 (1.9) 97.5 (1.7)
0.1 (0.1) 8.4 (2.6) 0.5 (0.4)
55.2 (10.4) 62.9 (12.2) 59.2 (12.6)
59.1 (5.6) 64.4 (10) 72.6 (12.1)
27.5 (6.7) 35.3 (11.5) 42 (18.6)
2.1 (0.9) 2.5 (1.1) 7.5 (4.4)
34.5 (2.7) 38 (3.8) 41.3 (6.2)
hts; WR, weighed records;
ied by number of completed recalls.
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338
FIGURE 1. Food groups consumed by WR, FRANI application, and 24HR (n ¼ 36). The mean number of food groups consumed was 6.63 for WR,
6.62 for FRANI, and 7.02 for 24HRs. FRANI, Food Recognition Assistance and Nudging Insights; WR, weighed records; 24HR, 24-hr recall.riboflavin (24HR). The CCCs by nutrient between FRANI andWR
ranged between 0.30 and 0.68, with similar CCCs found for
24HR and WR (ranging between 0.38 and 0.67) (Figure 3 and
Supplemental Table S2).
Sources of error
Examination of the percentage of energy intake consumed by
food group highlighted minor variations across FRANI and 24HR
compared with WR (Table 4). However, more substantive vari-
ations were found in the quantities of food consumed by food
groups. FRANI and 24HR methods appeared to underestimate
most quantities consumed of all food groups except for “meat,
poultry, and fish” and “eggs” food groups (and 24HR for “other
vitamin A-rich fruits and vegetables”). Tests for statistically sig-
nificant differences were not performed as the purpose of Table 4
is purely descriptive. Comparisons of food consumption episodes
from FRANI and WR found levels of omission and intrusion er-
rors of 31% and 16%, respectively. Omission and intrusion errors
were lower when comparing 24HR with WR (21% and 13%,
respectively). Further analysis of the most commonly consumed
foods showed no clear bias in terms of over- or underreporting of
consumption episodes or portion estimation for the different
methods (Table 5).2333Discussion
By comparing the estimates from FRANI with WR, the gold
standard for dietary assessment, and with 24HR, a standard
method commonly used for dietary assessment, this study pro-
vides new rigorous evidence on the relative validity of using
innovative AI–assisted mobile phone technology to assess the
diets of female adolescents in LMIC. The findings suggest that the
AI–assisted dietary assessment can estimate energy intake in
adolescent females within a 10% equivalence bound, whereas
the energy intake estimate from 24HR was accurate at a 20%
bound compared with WR. For FRANI, equivalence was within a
15% bound for folate, iron, zinc, niacin, and vitamin B6, and the
intake of 5 more micronutrients estimated within a 20% bound.
No estimates of micronutrient intake obtained from the 24HR
method were accurate within a 15% equivalence bound,
although intake of 3 micronutrients estimated through 24HR fell
within a 20% bound. Errors in MPA prevalence by method were
lower for FRANI (~10 and ~3 percentage points) than for 24HR
(~20 and ~6 percentage points). Although these differences are
relatively small in magnitude, they provide evidence that FRANI
is at least as accurate, and may be marginally more accurate than
human-assisted dietary assessment in this particular context.
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338
FIGURE 2. Equivalence testing of ratios of log-transformed nutrient intakes comparing (A) FRANI vs. WR and (B) 24HR vs. WR. FRANI, Food
Recognition Assistance and Nudging Insights; WR, weighed records; 24HR, 24-hr recall.These findings are consistent with a validation study of FRANI
with a very similar design conducted in female adolescents in
Thai Nguyen, a secondary city in Vietnam [24]. In that study,
both FRANI and 24HR performed relatively more efficiently than
in Ghana, with equivalence between FRANI and WR determined
at the 10% bound for energy, protein, and fat and 4 nutrients
(iron, riboflavin, vitamin B-6, and zinc), and at 15% and 20%
bounds for carbohydrate, calcium, vitamin C, thiamin, niacin,
and folate. Similar results were observed for differences between
24HRs and WR with a 20% equivalent bound for all nutrients
except for vitamin A. The differences observed in the perfor-
mance of FRANI could be explained by differences in the un-
derlying food and recipe databases. It is also possible that the
skills of enumerators performing the 24HR were different be-
tween the 2 countries. Further research is underway to examine2334the drivers of the underperformance of FRANI and 24HR in
Ghana relative to Vietnam. One driver may be related to the
socioeconomic differences in the 2 study populations. In Ghana,
the study population was randomly selected across all income
groups regardless of education status, whereas in Vietnam the
study sample was selected from adolescent females attending
high school.
We found wide bounds (>30%) for vitamin A (FRANI) and
vitamin B12 (FRANI, 24HR), and riboflavin (24HR). These were
likely due to a combination of estimation errors and large vari-
ance in the actual intake of these nutrients (partly due to the
small sample size). For these 3 nutrients, the frequency of con-
sumption of foods was low with extremely high nutrient content
including gizzard, “sobolo,” sardines, mackerel, herrings, and
eggs, leading to extreme values skewing the relevant nutrient
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338
FIGURE 3. Concordance correlation coefficients of nutrient intakes on 3 d by WR, FRANI application, and 24HR (n ¼ 36). The concordance
correlation coefficient was estimated with adjustment for repeated measures. FRANI, Food Recognition Assistance and Nudging Insights; RAE,
retinol activity equivalents; WR, weighed records; 24HR, 24-hr recall.
TABLE 4
Quantity of food consumed by major food group and share of energy intake from major food group over 3 d by WRs, FRANI application, and 24HR
recall
Quantity consumed (g) WR (n ¼ 99)1 FRANI (n ¼ 99)1 24HR (n ¼ 99)1
Mean (SD) Median Mean (SD) Median Mean (SD) Median
Grains, white roots and tubers, and plantains 624.1 (324.1) 552.0 465.0 (323.2) 381.6 619.2 (355.3) 606.6
Pulses (beans, peas, and lentils) 102.2 (145.9) 0.0 76.2 (123.5) 0.0 99.12(150.6) 0.0
Nuts and seeds 40.5 (96.1) 0.0 30.6 (84.4) 0.0 32.6 (87.3) 0.0
Dairy 75.0 (126.3) 0.0 30.8 (55.8) 0.0 69.1 (114.7) 9.0
Meat, poultry, and fish 279.7 (209.0) 286.0 279.1 (223.8) 222.0 301.1 (227.6) 272.0
Eggs 42.3 (80.0) 0.0 47.7 (72.7) 9.5 64.1 (91.3) 0.0
Dark green leafy vegetables 9.7 (28.0) 0.0 6.3 (25.6) 0.0 9.5 (28.2) 0.0
Other vitamin A-rich fruits and vegetables 369.1 (205.9) 361.0 337.4 (236.8) 293.0 372.0 (230.4) 370.8
Other vegetables 402.1 (259.1) 399.0 370.1 (284.5) 315.4 434.4 (276.2) 381.0
Other fruits 59.0 (120.5) 0.0 8.1 (23.0) 0.0 49.2 (109.2) 0.0
Share of energy intake Mean (SD) Median Mean (SD) Median Mean (SD) Median
Grains, white roots and tubers, and plantains 0.2 (0.1) 0.2 0.2 (0.16) 02 0.2 (0.1) 0.2
Pulses (beans, peas, and lentils) 0.0 (0.1) 0.0 0.0 (0.06) 0.0 0.0 (0.1) 0.0
Nuts and seeds 0.0 (0.0) 0.0 0.0 (0.05) 0.0 0.0 (0.0) 0.0
Dairy 0.1 (0.1) 0.0 0.0 (0.09) 0.0 0.1 (0.1) 0.0
Meat, poultry, and fish 0.1 (0.1) 0.1 0.1 (0.12) 0.1 0.1 (0.1) 0.1
Eggs 0.0 (0.1) 0.0 0.1 (0.08) 0.0 0.0 (0.1) 0.0
Dark green leafy vegetables 0.0 (0.0) 0.0 0.0 (0.0) 0.0 0.0 (0.1) 0.0
Other vitamin A-rich fruits and vegetables 0.2 (0.1) 0.2 0.3 (0.2) 0.2 0.2 (0.1) 0.2
Other vegetables 0.1 (0.1) 0.1 0.2 (0.1) 0.1 0.1 (0.1) 0.1
Other fruits 0.0 (0.1) 0.0 0.0 (0.0) 0.0 0.0 (0.1) 0.0
FRANI, Food Recognition Assistance and Nudging Insights; WR, Weighed Records
1 Number of person-days ¼ 99, equal to number of subjects multiplied by number of completed recalls.intake distribution. Concerning the estimation errors for vitamin
A intake in FRANI, lowering the standard portion size in the
database on portion sizes for manual entry of “sobolo” (hibiscus
drink) would likely increase the overall estimation accuracy and
reduce the large variance observed.
To our knowledge, no other studies assess the validity of
mobile technology for dietary assessment in adolescents in LMIC.
A systematic review including 14 validation studies of mobile
phone–based applications in high-income countries with 22335focusing on adolescents in Sweden [32] and Korea [33] exam-
ined the performance of mobile applications relative mostly to
24HR, with only 2 comparing with WR [34]. The analysis found
that dietary record applications slightly underestimated food
consumption compared with traditional dietary assessment
methods (–85 kcal/d for energy;19 g/d,13 g/d, and12 g/d
for carbohydrate, fat, and protein intake, respectively). The ev-
idence presented in this validation study is thus consistent with
that reported in the emerging literature on this topic,
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338
TABLE 5
Comparisons of reported consumption episodes and mean quantities eaten among most commonly consumed foods assessed using WR, FRANI, and
24HR
Food Consumption episodes (n) Mean quantity consumed (g) Ratio of quantities consumed
WR FRANI 24HR WR FRANI 24HR FRANI/WR 24HR/WR
Shito 35 46 34 34 35 18 1.05 0.52
Banku 25 26 22 364 338 245 0.93 0.67
Tomato stew 25 24 23 52 35 52 0.67 0.99
Chicken fried 24 33 32 45 80 103 1.80 2.30
Rice jollof 24 25 25 286 249 271 0.87 0.95
Spaghetti 24 27 21 226 145 110 0.64 0.49
Egg boiled 23 27 22 48 71 60 1.46 1.24
Bread tea 22 31 15 97 126 324 1.31 3.36
Fish fried 20 16 19 59 57 67 0.97 1.13
Hausa Koko 17 22 16 353 204 454 0.58 1.28
Rice plain boiled 17 24 21 318 207 301 0.65 0.95
Waakye 17 20 16 299 168 345 0.56 1.15
Milk 15 1 33 33 44 31 1.35 0.95
FRANI, Food Recognition Assistance and Nudging Insights; WR, Weighed Records; 24HR, 24-hr recall.highlighting the potential for new technologies to improve diet
assessment methods and potentially reduce data collection costs.
A major strength of this analysis is the rigorous design of the
validation study, including comparisons with both WR (as the
gold standard for dietary assessment) and 24HR (the standard
method commonly used in studies), and data collected by
different enumerators across these 2 methods to reduce potential
bias. The design included data collection on 3 nonconsecutive
days, including week and weekend days, to provide a perspective
on usual intake. Nutrient intakes were computed from a compi-
lation of locally relevant food-composition databases and the
West African Food-Composition Tables within which a majority
of foods were identified. In the few cases where exact foods could
not be found, nutrient values of similar foods were substituted.
Several limitations, however, are also apparent. First, the
study was conducted in a relatively controlled environment,
where FRANI use could be to some degree assisted by WR enu-
merators. This could lead to higher precision in FRANI estima-
tion compared with that expected in real-world use. In parallel,
the precision of the 24HR could potentially also be biased toward
higher accuracy as participants were likely primed during the
recall by FRANI use on the previous day, suggesting that the
overall comparisons between FRANI and 24HR presented here
may also hold in a real-world setting. When 24HRs do not ac-
count for all foods and beverages consumed, this would lead to
less precision on the nutrient intakes and wider confidence
bounds. This topic remains an important area of ongoing
research. In addition, although the sample size was small, it was
in line with expectations for a pilot study, and participants were
recruited randomly from a population representative of adoles-
cents in Accra, thus increasing the external validity of these re-
sults. In addition, due to challenges in matching records across
the 3 methods, the analytical sample included 98 out of the 108
records resulting in wider CIs. However, the degree of acceptable
error is open for debate and should be carefully considered in the
context of the dietary information required. In this study,
although the 15% bound for energy is equal to approximately
280 kcals, which is similar or better than some validation studies
in adults in the literature, this level of precision may be consid-
ered unacceptable when high degrees of accuracy are required.2336Finally, although the expectation for the WRwas to cover the full
24-h period, for logistic reasons enumerators were personally
only covering the period from early morning (~6 am) to late
evening (after last meal was supposedly consumed). In this
analysis, we matched consumption times across the 3 methods,
excluding food consumption reported outside the WR period.
In conclusion, we found that FRANI AI–assisted dietary
assessment could accurately estimate nutrient intakes in
adolescent females compared with WR in urban Ghana. FRANI
estimates were at least as accurate as those provided through
24HR. Further improvements in food recognition and portion
estimation in FRANI could reduce errors and improve overall
nutrient intake estimations. Further research is ongoing,
including pilots on the feasibility of using FRANI to nudge ado-
lescents toward healthy food choices in LMICs. Although the
potential for new technologies such as FRANI to have an impact
at scale is clear, real-world evaluations are needed to ensure that
the technology is feasible, valid, and effective.
Author contributions
A.G., G.F., and P.N. designed the research; P.M., R.G., F.D.,
and D.H. developed the FRANI application; P.H.N., L.M.T.,
N.T.H., B.K., P.M., R.G., G.F., B.B., A.A., B.C.B., J.A., A.K., F.D.,
D.H., and A.G. conducted the research; A.G. and G.F. analyzed
and interpreted the data; G.F., A.G., and P.N. wrote the paper;
G.F., P.H.N., A.G., J.A., B.K., and P.M. provided the critical
revision of the manuscript for important intellectual content. All
authors have read and approved the final manuscript.
Data Availability
Data described in the manuscript, code book, and analytic
code will be made available upon request.
Funding
Funding for this research was provided by Foundation Botnar
and the CGIAR program on Agriculture for Nutrition and Health
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338(A4NH). The funders had no role in study design, data collection
and analysis, data interpretation, the decision to publish, or
preparation of the manuscript.
Author disclosures
G.F., B.B., V.A., G.A., B.K., P.M., R.G., A.A., B.C.B., J.A., A.K.,
F.D., L.M.T., N.T.H., D.H., P.H.N., and A.G. have no conflicts of
interest.
Acknowledgments
We are also grateful to our study participants and the Accra
Adolescent study team, including Jef Leroy, Elodie Becquey,
Lieven Huybregts, Richmond Aryeetey, and Rock Zagre for
sharing the data on adolescents for our sampling, and for their
support.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http
s://doi.org/10.1016/j.tjnut.2023.06.001.
References
[1] A. Afshin, P.J. Sur, K.A. Fay, L. Cornaby, G. Ferrara, J.S. Salama, et al.,
Health effects of dietary risks in 195 countries, 1990–2017: a systematic
analysis for the Global Burden of Disease Study 2017, Lancet 393
(10184) (2019) 1958–1972, https://doi.org/10.1016/S0140-6736(19)
30041-8.
[2] J.M. Saavedra, The changing landscape of children’s diet and nutrition:
new threats, new opportunities, Ann. Nutr. Metab. 78 (Suppl 2) (2022)
40–50, https://doi.org/10.1159/000524328.
[3] B.M. Popkin, C. Corvalan, L.M. Grummer-Strawn, Dynamics of the
double burden of malnutrition and the changing nutrition reality,
Lancet 395 (10217) (2020) 65–74, https://doi.org/10.1016/S0140-
6736(19)32497-3.
[4] A. Must, R.S. Strauss, Risks and consequences of childhood and
adolescent obesity, Int. J. Obes (Lond). 23 (2) (1999) S2–S11, https://
doi.org/10.1038/sj.ijo.0800852.
[5] F.M. Biro, M. Wien, Childhood obesity and adult morbidities, Am. J.
Clin. Nutr. 91 (5) (2010) 1499S, https://doi.org/10.3945/
ajcn.2010.28701B, 505S.
[6] D.S. Freedman, W.H. Dietz, S.R. Srinivasan, G.S. Berenson, The relation
of overweight to cardiovascular risk factors among children and
adolescents: the Bogalusa Heart Study, Pediatrics 103 (6) (1999)
1175–1182, https://doi.org/10.1542/peds.103.6.1175.
[7] A. Must, Morbidity and mortality associated with elevated body weight
in children and adolescents, Am. J. Clin. Nutr. 63 (3) (1996)
445S–447S, https://doi.org/10.1093/ajcn/63.3.445.
[8] T. Bjørge, A. Engeland, A. Tverdal, G.D. Smith, Body mass index in
adolescence in relation to cause-specific mortality: a follow-up of
230,000 Norwegian adolescents, Am. J. Epidemiol. 168 (1) (2008)
30–37, https://doi.org/10.1093/aje/kwn096.
[9] A. Peeters, J.J. Barendregt, F. Willekens, J.P. Mackenbach, A.A. Mamun,
L. Bonneux, et al., Obesity in adulthood and its consequences for life
expectancy: a life-table analysis, Ann. Intern. Med. 138 (1) (2003)
24–32, https://doi.org/10.7326/0003-4819-138-1-200301070-00008.
[10] D. Hargreaves, E. Mates, P. Menon, H. Alderman, D. Devakumar,
W. Fawzi, et al., Strategies and interventions for healthy adolescent
growth, nutrition, and development, Lancet 339 (10320) (2021)
P198–P210, https://doi.org/10.1016/S0140-6736(21)01593-2.
[11] E.C. Keats, A.I. Rappaport, S. Shah, C. Oh, R. Jain, Z.A. Bhutta, The
dietary intake and practices of adolescent girls in low- and middle-
income countries: a systematic review, Nutrients 10 (12) (2018) 1978,
https://doi.org/10.3390/nu10121978.
[12] R. Kupka, K. Siekmans, T. Beal, The diets of children: overview of
available data for children and adolescents, Glob. Food Sec. 27 (2020)
100442, https://doi.org/10.1016/j.gfs.2020.100442.2337[13] SPRING, Adolescent nutrition. Call to action: better data now to drive
better policies and programs in the future Arlington. VA Strengthening
Parternships, Results, and Innovations in Nutrition Globally (SPRING)
Project, 2018. Available from: https://www.spring-nutrition.org/about
-us/news/adolescent. (Accessed 3 July 2023).
[14] J.E. Arsenault, M. Moursi, D.K. Olney, E. Becquey, R. Ganaba,
Validation of 24-h dietary recall for estimating nutrient intakes and
adequacy in adolescents in Burkina Faso, Matern, Child Nutr 16 (4)
(2020), e13014, https://doi.org/10.1111/mcn.13014.
[15] M. Livingstone, P. Robson, Measurement of dietary intake in children,
Proc. Nutr. Soc. 59 (2) (2000) 279–293, https://doi.org/10.1017/
s0029665100000318.
[16] J.L. Fiedler, Y. Martin-Pr
evel, M. Moursi, Relative costs of 24-hour
recall and household consumption and expenditures surveys for
nutrition analysis, Food Nutr. Bull. 34 (3) (2013) 318–330, https://
doi.org/10.1177/156482651303400304.
[17] W. Bell, B.A. Colaiezzi, C.S. Prata, J.C. Coates, Scaling up dietary data
for decision-making in low-income countries: new technological
frontiers, Adv. Nutr. 8 (6) (2017) 916–932, https://doi.org/10.3945/
an.116.014308.
[18] B.C. Braga, N.-L. Aberman, A. Arrieta, B. Bannerman, A. Burns,
G. Folson, et al., Design of a mobile phone-based artificial intelligence
(AI) application to assess dietary intake and provide nudges to improve
healthy eating choices: formative research in Ghana and Vietnam, Int.
Food Policy Res. Inst. (2021). https://doi.org/10.2499/p15738coll
2.134412. (Accessed 3 July 2023).
[19] M.M. Jankowska, J.R. Weeks, R. Engstrom, Do the most vulnerable
people live in the worst slums? A spatial analysis of Accra, Ghana, Ann
GIS 17 (4) (2011) 221–235, https://doi.org/10.1080/
19475683.2011.625976.
[20] J.R. Weeks, A. Hill, D. Stow, A. Getis, D. Fugate, Can we spot a
neighborhood from the air? Defining neighborhood structure in Accra,
Ghana, GeoJournal. 69 (1) (2007) 9–22, https://doi.org/10.1007/
s10708-007-9098-4.
[21] C.N. Tugault-Lafleur, J.L. Black, S.I. Barr, A systematic review of
methods to assess children's diets in the school context, Adv. Nutr. 8 (1)
(2017) 63–79, https://doi.org/10.3945/an.116.013144.
[22] M.J. Batterham, C. Van Loo, K.E. Charlton, D.P. Cliff, A.D. Okely,
Improved interpretation of studies comparing methods of dietary
assessment: combining equivalence testing with the limits of
agreement, Br. J. Nutr. 115 (7) (2016) 1273–1280, https://doi.org/
10.1017/S0007114516000040.
[23] F.E. Thompson, S. Dixit-Joshi, N. Potischman, K.W. Dodd,
S.I. Kirkpatrick, L.H. Kushi, et al., Comparison of interviewer-
administered and automated self-administered 24-hour dietary recalls
in 3 diverse integrated health systems, Am. J. Epidemiol. 181 (12)
(2015) 970–978, https://doi.org/10.1093/aje/kwu467.
[24] P.H. Nguyen, L.M. Tran, N.T. Hoang, D.T.T. Trương, T.H.T. Tran,
P.N. Huynh, et al., Relative validity of a mobile AI-technology–assisted
dietary assessment in adolescent females in Vietnam, Am. J. Clin. Nutr.
116 (4) (2022) 992–1001, https://doi.org/10.1093/ajcn/nqac216.
[25] L. Lin, A.S. Hedayat, W. Wu, A unified approach for assessing agreement
for continuous and categorical data, J. Biopharm. Stat. 17 (4) (2007)
629–652, https://doi.org/10.1080/10543400701376498.
[26] R.S. Gibson, E.L. Ferguson, An interactive 24-hour recall for assessing
the adequacy of iron and zinc intakes in developing countries.
HarvestPlus Technical Monograph 8, in: International Food Policy
Research Institute (IFPRI) and International Center for Tropical
Agriculture, CIAT), Washington, DC and Cali, 2008.
[27] A. Vincent, F. Grande, E. Compaor
e, G. Amponsah Annor, P.A. Addy,
L.C. Aburime, et al., FAO/INFOODS food composition table for Western
Africa (2019) user guide & condensed food composition table [Table de
composition des aliments FAO/INFOODS pour l’Afrique de l’Ouest
(2019) Guide d’utilisation & table de composition des aliments
condens
ee], FAO, Rome, 2020. https://www.fao.org/3/ca7779b/CA
7779B.PDF. (Accessed 3 July 2023).
[28] B. Bannerman, R. Soueida, A. Ohemeng, RIING Nutrient Database: a
compilation of nutrient compositions of various foods under the research
to improve infant nutrition and growth project (2003–2008), Department
of Nutrition and Food Science, University of Ghana, Legon, Ghana, 2018.
[29] European Food Safety Authority, Dietary reference values for nutrients
summary report, EFSA Supporting Publications 14 (12) (2017)
e15121E. Available from: https://efsa.onlinelibrary.wiley.com/doi/epd
f/10.2903/sp.efsa.2017.e15121.
[30] Food and Agriculture Organization, Family Health International 360,
Minimum dietary diversity for women: a guide for measurement, FAO,
G.K. Folson et al. The Journal of Nutrition 153 (2023) 2328–2338Rome, 2016. https://www.fao.org/3/i5486e/i5486e.pdf. (Accessed 3
July 2023).
[31] H.X. Barnhart, M.J. Haber, L.I. Lin, An overview on assessing agreement
with continuous measurements, J. Biopharm. Stat. 17 (4) (2007)
529–569, https://doi.org/10.1080/10543400701376480.
[32] Å. Svensson, C. Larsson, A mobile phone app for dietary intake
assessment in adolescents: an evaluation study, JMIR mhealth uhealth 3
(4) (2015), e4804, https://doi.org/10.2196/mhealth.4804.2338[33] J.-E. Lee, S. Song, J.S. Ahn, Y. Kim, J.E. Lee, Use of a mobile application
for self-monitoring dietary intake: feasibility test and an intervention
study, Nutrients 9 (7) (2017) 748, https://doi.org/10.3390/
nu9070748.
[34] L. Zhang, A. Misir, H. Boshuizen, M. Ock
e, A systematic review and
meta-analysis of validation studies performed on dietary record apps,
Adv. Nutr. 12 (6) (2021) 2321–2332, https://doi.org/10.1093/
advances/nmab058.