A machine learning approach to using ultrasound for body composition and nutritional status assessment in newborns: a pilot study protocol

Abstract

Background

Accurate assessment of infant body composition, specifically fat and fat-free mass, is crucial for evaluating growth and nutritional status. Existing methods, such as air displacement plethysmography and dual-energy X-ray absorptiometry, are expensive, require specialized facilities, and demand trained personnel. Ultrasound offers a promising alternative as a portable, low-cost tool capable of distinguishing adipose tissue from skeletal muscle. However, its feasibility for widespread use in diverse clinical settings remains uncertain. This pilot study aims to assess the feasibility, acceptability, and reliability of an ultrasound scanning protocol for measuring body composition in infants while establishing a foundation for artificial intelligence (AI)-enabled analysis to enable whole-body composition estimation from ultrasound images.

Methods

We will recruit 50 infants (20 preterm and 30 full-term infants) from two sites: Brigham and Women’s Hospital (Boston, MA, USA) and Jimma Medical Center (Jimma, Ethiopia). Feasibility will be assessed through metrics such as recruitment rates, scan completion, and session duration. Acceptability will be measured using clinician feedback, and reliability will be evaluated using intra-class correlation coefficients (ICCs) for ultrasound image acquisition. A comprehensive database of ultrasound images and corresponding body composition metrics will be developed, forming the foundation for training AI models. Preliminary machine learning (ML) models, including convolutional neural networks (CNN), will be developed to predict body composition. The accuracy of these models will be evaluated using metrics such as root mean squared error (RMSE), mean absolute percent error (MAPE), mean absolute error (MAE), and mean squared error (MSE).

Discussion

This study will determine the feasibility of integrating ultrasound-based body composition assessment into neonatal clinical workflows as a potential future application. We will evaluate protocol adherence, scan reliability, and clinician and family acceptability to guide further protocol optimization. Findings will inform the design of a larger-scale study and contribute to refining AI models for clinical use. Ultimately, this approach aims to improve the accessibility, accuracy, and efficiency of body composition assessments, particularly in low-resource settings, where it could enable frontline healthcare workers to perform these assessments without specialized training, improving care for vulnerable infants.

Background

Global malnutrition remains prevalent and is linked to significant global health challenges, often described as a “double burden” in populations where both undernutrition and overnutrition coexist [1]. The issue has profound consequences that affect countries regardless of their economic status. For instance, early-life undernutrition leads to conditions such as stunting and wasting, contributing to nearly 50% of deaths among children under 5 years old globally [2]. Survivors often face increased susceptibility to disease, diminished physical performance, and challenges in educational advancement [2]. Later in life, obesity becomes a primary driver of chronic non-communicable diseases, including diabetes, heart disease, and hypertension [3].

Traditionally, healthcare providers monitor nutritional health through anthropometric indicators such as weight, height, and body mass index (BMI). While these measurements are relatively easy to obtain and provide valuable information, they do not provide a comprehensive understanding of healthy growth and development, often overlooking critical aspects such as the distribution of body fat and lean mass. This is particularly true for infants, in whom BMI is a poor indicator of adiposity [4, 5]. For example, a newborn may regain weight following treatment for malnourishment; however, without additional measures, it remains unclear whether this weight gain stems from unhealthy fat accumulation or healthy muscle development [6, 7]. As a result, reliance solely on anthropometric indicators can lead to misinterpretations of an individual's nutritional status, potentially hindering the implementation of targeted and effective interventions.

Recently, there has been increased interest in employing body composition measures, such as fat mass (FM) and fat-free mass (FFM), to assess growth quality and nutrition accretion, as a means to more effectively target nutritional interventions [8]. Despite promising findings, obtaining body composition measurements remains challenging due to costly equipment and the need for specialized facilities and trained personnel. Such barriers significantly limit their applicability, particularly at the point of care and in low-resource settings. Ultrasound imaging offers a promising, cost-effective method for evaluating adipose tissue and skeletal muscle, positioning it as a scalable tool for assessing nutritional status and guiding appropriate interventions [9, 10]. The growing availability of portable ultrasound probes further enhances its accessibility as a point of care tool [11].

Several recent studies have demonstrated the feasibility of using ultrasound to assess regional tissue characteristics in neonatal and infant populations. Early work showed that ultrasound measurements of subcutaneous fat and muscle thickness could be obtained at the bedside in preterm infants; however, the reliability of thickness-based measurements was variable and depended strongly on operator technique and image quality [12]. Some more recent studies have reported improved feasibility and reliability of ultrasound-based tissue assessment in hospitalized infants evaluated for malnutrition and medically fragile very preterm infants, which support the practicality of standardized imaging protocols in neonatal care settings, but these studies were performed in high-resource settings and relied on ultrasound devices that may not be accessible in all practice settings [13, 14]. In parallel, studies from our group have demonstrated the feasibility of using regional ultrasound images as inputs to deep learning models to estimate whole-body composition, including FM and FFM, in newborn populations [15, 16]. The present study builds on prior work by developing a neonatal imaging protocol that incorporates standardized video sweep acquisition to reduce user variability, implemented with a low-cost portable ultrasound system, and evaluating feasibility across diverse clinical settings.

In this pilot study, ultrasound is used to acquire images and video sweeps at specific anatomical regions in a cohort of hospitalized infants. This study aims to assess the feasibility, acceptability, and inter-rater reliability of a low-cost, portable ultrasound device for regional image acquisition in neonates. These images will then serve as inputs for exploratory AI-based analysis to predict whole-body composition, including FM and FFM, obtained using air displacement plethysmography (ADP) as the reference standard. This pilot study will inform the development of a larger-scale study, with the ultimate goal of developing AI models to enable body composition prediction from ultrasound image analysis.

Study aims

The ultimate objective of this research is to develop and validate a novel artificial intelligence (AI)-enabled ultrasound tool for measurement of critical body composition metrics in infants, with the aim of improving nutritional assessment and clinical outcomes. The tool will leverage machine learning (ML) models to predict body composition and will be designed for use by frontline healthcare workers, including those with minimal training in ultrasound imaging. This pilot study will establish feasibility and build the foundation for a larger-scale study by addressing the following objectives:

Aim 1

Evaluate the feasibility, acceptability, and reliability of an ultrasound image and video acquisition protocol across diverse neonatal care settings. We will assess the feasibility of implementing the ultrasound protocol across different care environments, including well-baby nurseries, neonatal intensive care units (NICUs), and both low- and high-resource settings. Metrics for feasibility will include protocol adherence, scanning success rates, and operational efficiency. Acceptability will be evaluated through feedback from healthcare providers and families, focusing on usability and perceived benefits. Reliability will be measured through inter-rater consistency in image acquisition.

Aim 2

Establish a comprehensive database of ultrasound images and corresponding whole-body composition metrics from ADP. We will refine and finalize a clinical protocol for consistent ultrasound data collection and gather high-quality ultrasound images alongside gold-standard measures of body composition obtained via ADP. This database will serve as the foundation for subsequent model development and validation.

Secondary aim

Pilot the development of ML models for predicting infant whole-body composition and nutritional status from regional ultrasound images. Using the data collected in Aim 2, we will process and annotate ultrasound images with “ground truth” measures of body composition. Preliminary ML models, including convolutional neural networks (CNN), will be trained to identify key image markers predictive of body composition and to evaluate the utility of video sweeps in reducing operator-dependent variability. We will also explore the accuracy and generalizability of these models, utilizing interpretability tools to ensure alignment with clinical landmarks.

Methods

Study design

This multi-center, prospective observational cohort study evaluates the feasibility, acceptability, and reliability of using ultrasound technology to evaluate body composition in full-term and preterm infants. Conducted at clinical sites in Boston and Ethiopia, this study will examine outcomes across diverse healthcare settings. Data will be collected at multiple time points: once within 72 h for full-term infants and periodically for preterm infants once clinically stable. This design ensures comprehensive evaluation of the ultrasound tool’s utility for clinical practice and model development. As a pilot feasibility study, this investigation is not designed to test clinical hypotheses or validate predictive models, but rather to evaluate implementation feasibility and inform future larger-scale studies.

Study setting

Data will be collected at two primary sites: one in the U.S. and one in Ethiopia. In the USA, the study will take place at the Brigham and Women’s Hospital, a level 3 NICU and the largest high-risk delivery service in Massachusetts, along with the associated well-baby/Postpartum Unit. In Ethiopia, the data collection will occur in the NICU and Outpatient Pediatric Department (OPD) of Jimma Medical Center, a tertiary referral hospital located in the southwest region of the country. These diverse settings will enable the evaluation of the feasibility of using ultrasound in both high- and low-resource settings.

Differences between study environments may influence implementation feasibility. These include variations in staffing structures, availability of ultrasound equipment and supporting technology, clinical workflow constraints, and patient population characteristics. For example, differences in provider experience, bedside workflow demands, and resource availability may affect scan timing, protocol adherence, and data collection efficiency. These contextual factors will be documented and considered when interpreting feasibility outcomes.

Study population

Inclusion criteria require infants to maintain stable cardiovascular and respiratory status without the need for intensive support at the time of measurement. Infants will be excluded if they have congenital anomalies of the limbs, injuries, or medical conditions that would interfere with ultrasound performance. Additionally, infants with genetic anomalies affecting growth or nutrient accretion will be excluded. By strictly enrolling infants who can safely undergo the ADP reference standard, we ensure a high-quality "ground truth" for the subsequent development of the AI-enabled ultrasound models.

Clinically stable full-term and preterm infants will be recruited from the well-baby nursery or NICU during their initial hospitalization after birth. Full-term infants will undergo a single measurement within the first 72 h after birth, prior to discharge. Preterm infants (born at less than 37 weeks gestation) will be enrolled once deemed clinically stable by the primary medical team. Because the ADP device requires an infant to be free of external medical equipment (e.g., respiratory support or intravenous lines), final eligibility for each measurement is confirmed immediately prior to the procedure. For preterm infants, this assessment occurs once they have been weaned off respiratory support, allowing for longitudinal measurements at intervals of 1 to 2 weeks as the length of hospital stay permits until hospital discharge.

Outcome measures

Outcome measures will be categorized into four groups: feasibility, reliability, and acceptability; database creation; influence of clinical descriptor on body composition; and ML model validation. Feasibility, acceptability, and reliability outcomes will be summarized both overall and stratified by study site to evaluate protocol performance across high-resource and low-resource clinical settings. A summary of the outcome measures is provided in Table 1.

Table 1.

Outcome measures and data collection methods

Concept/Variable	Measures
Feasibility outcomes	Enrollment rate of eligible infants, percentage of successful scans, average time per ultrasound session
Reliability outcome	Inter-rater reliability of ultrasound measurements
Acceptability outcomes	Clinician surveys assessing perceived ease of use, satisfaction, and barriers encountered
Database of annotated ultrasound images and body composition metrics	Ultrasound method: 3 × image and video sweep at biceps (brachii and brachialis), abdomen (rectus abdominis), and quadriceps (rectus femoris and vastus intermedius)
	Air-displacement plethysmography method (Peapod): FM, FFM, body volume, body density
	Anthropometric measures: length, weight, mid-upper arm circumference, head circumference
Clinical descriptors	History of Hospitalization form and feeding data collection form for secondary clinical variables: infant age, health condition, sex, race/ethnicity and dietary intake (e.g., breastmilk or formula)
Validation of ML algorithms for body composition estimation	Accuracy of models will be assessed using metrics such as mean squared error (MSE) and R-squared (R2)

Feasibility, reliability, and acceptability outcomes

Feasibility will be assessed through several metrics such as the percentage of eligible infants enrolled, the percentage of successful scans completed, and the average time required for each ultrasound session. Feasibility will be considered successful if the following thresholds are met: 80% approach of eligible infants, 90% completion of scheduled scans, and the average ultrasound session time of no more than 30 min. These feasibility metrics will also be used to evaluate implementation of the novel ultrasound scanning protocol elements, including the use of handheld ultrasound devices, video sweep acquisition, and clinical workflow integration across different clinical settings. Inter-rater reliability will be evaluated by comparing ultrasound scans taken by different clinicians on the same infant. Consistency across operators will be evaluated to ensure consistency of standardized image acquisition and anatomical landmark visualization. Rather than focusing on manual caliper measurements, this assessment targets the operator’s ability to reliably capture the specific anatomical planes and tissue interfaces required for downstream automated AI analysis. Reliability will be considered acceptable if the ICC exceeds 0.80, which is considered indicative of good reliability in image acquisition across different clinicians [17]. Reliability assessments will include evaluation of image quality and consistency across both still image and video sweep acquisitions, with specific focus on the visibility of the fat-muscle interface and underlying bone markers. Acceptability will be evaluated through clinician surveys, focusing on the ease of use, satisfaction with the technology, and barriers encountered during implementation. Acceptability will be considered satisfactory if 75% of clinicians report ease of use and satisfaction with the technology, with no major barriers during data collection.

Database of annotated ultrasound images and body composition metrics

An outcome of the pilot study will be the creation of a database containing ultrasound images and videos of muscle and adipose tissue in both full-term and preterm infants. These images will be annotated with “gold standard” measurements of body composition (FM and FFM) derived from ADP, along with corresponding anthropometric measures, such as weight, length, head circumference, and mid-upper arm circumference. This database will serve as a critical resource for subsequent ML model development. The database will be considered complete if at least 90% of images and videos pass quality checks, including appropriate imaging depth, acceptable compression levels, and absence of significant artifacts.

Influence of clinical descriptors on body composition

Secondary clinical variables, such as infant age, health condition, sex, race/ethnicity, and dietary intake (e.g., breastmilk or formula), will be collected to evaluate their potential influence on body composition. These variables will be incorporated as covariates in the ML models to assess their contribution to the prediction of body composition and nutritional status. Analysis of clinical descriptors will be strictly exploratory and hypothesis generating; these data will be used to identify potential covariates for future, adequately powered validation studies.

Initial development of ML algorithms for body composition estimation

The expected outcome is the initial development and validation of ML algorithms using CNNs to predict whole-body composition measures, such as FM and FFM, from regional ultrasound image and video data. The accuracy of these models will be assessed in exploratory models using metrics such as (1) root mean squared error (RMSE), (2) mean absolute percentage error (MAPE), (3) mean absolute error (MAE), and (4) mean squared error (MSE). Our prior work demonstrated the feasibility of predicting neonatal whole-body composition using deep learning models trained on static ultrasound images collected in a single clinical setting [15, 16]. In contrast, the present pilot study focuses on establishing the feasibility of a standardized multisite data acquisition protocol using handheld ultrasound devices and incorporating video sweep.

Sample size

A total of 50 infants (25 from each site) will be enrolled, 10 preterm and 15 full-term infants at each location. This sample size was carefully selected to ensure sufficient precision for key feasibility outcomes, as well as to provide preliminary data for secondary outcomes, including model development. The enrollment rate for the study is estimated with a margin of error of approximately ± 15. In addition, the sample size is sufficient to estimate the scan completion rates with a precision of ± 12%, allowing for reliable assessment of adherence to the protocol. For estimating session duration, the sample size is sufficient to achieve an estimate within ± 5 min, with a standard deviation of 10 min at a 95% confidence level. The level of precision is appropriate for pilot data and ensures a reliable estimate of the average session duration, which will inform future study designs. Furthermore, the proposed sample size will allow for preliminary characterization of variability in body composition measurements, including fat mass, fat-free mass, body volume, and body density, which will help inform the design of a future larger study. This variability is not intended to support hypothesis testing, but rather to provide insight into how body composition metrics may differ based on infant-specific factors [18]. Although model development is a secondary aim of the study, analysis involving machine learning in this pilot study will be exploratory and intended to establish data processing pipelines and generate preliminary estimates, rather than to validate predictive performance. These calculations demonstrate that the sample size is suitable for the study’s feasibility objectives and will provide valuable data for the design of a larger, future study.

Data collection procedures

Anthropometric measures

Length will be measured by research staff to the nearest 0.1 cm using a length board (Ellard Instrumentation Ltd., Monroe, WA, USA) using the 2-person method [12]. Head circumference and mid-upper arm circumference are measured using a non-stretchable tape according to standard methods [19]. Length, head, and mid-upper arm circumference are each measured in duplicate and averaged to improve precision. Weight is measured to the nearest 0.1 g in the ADP device.

ADP

The Peapod Infant Body Composition System (COSMED, Concord, CA, USA) is a commercially available infant-specific device that uses ADP to measure body mass and volume, then uses densitometry principles to calculate whole-body fat and fat-free mass [20]. The accuracy, precision, and reproducibility of this device have been validated in both preterm and full-term infants weighing up to 8 kg [21, 22].

Clinical data

At the time of discharge, clinical data will be extracted from the infant’s medical chart, including demographics, birth and health history during hospitalization, and maternal pregnancy history. All data entered by site clinicians will be anonymized and de-identified before being entered into REDCap. Gestational age at birth, postnatal age, and post-menstrual age (PMA) at each ultrasound and body composition measurement time point will be recorded for all participants to account for the primary biological drivers of tissue development and maturation. For preterm infants, differences in assessment timing driven by clinical stability or hospital discharge—including potential discharge prior to term equivalent age—will be systematically documented and considered when interpreting feasibility and data collection completeness.

Standard operating procedures (SOP) for ultrasound data collection

A SOP for collecting ultrasound images and videos to assess body composition was developed based on published literature and consultations with pediatric radiologists [23, 24]. For imaging, the Clarius L20 and Clarius L15 high-frequency linear array scanners will be utilized, accompanied by a standard tablet device [25]. Handheld portable ultrasound devices were selected to facilitate bedside imaging across diverse clinical environments. Their use is intended to support workflow integration, reduce equipment constraints, and facilitate scalable implementation in both high-resource and low-resource settings. Feasibility outcomes will include evaluation of scan completion rates, session duration, and clinician workflow using these portable devices.

Measurements will be taken in triplicate on the right side of the infant's body, unless otherwise specified, once deemed clinically stable by the primary medical care team. The ultrasound probe will be placed perpendicular to the muscle of interest. To ensure that images can be effectively used for both AI training and clinical evaluation, it is crucial that skin, fat, muscle, and tissue regions are clearly distinguishable.

Zero or minimal compression will be applied during measurements to optimize visualization of the body's natural curvature. This technique is essential for accurately capturing anatomical features, as excessive pressure can distort soft tissues and compromise the appearance of muscle and adipose tissue. Operators will be trained to use a standardized minimal-contact technique, including application of a generous gel layer to allow the probe to rest on the skin surface without visible tissue deformation. Image quality review will include visual assessment for signs of compression, such as flattening of subcutaneous tissue layers, to ensure protocol adherence. Feedback will be provided to the data collection teams periodically to address any deviations regarding probe positioning, compression, or image clarity.

The anatomical placement of the ultrasound probe is outlined in Fig. 1, which depicts the three distinct muscle regions of interest (abdomen, biceps, and quadriceps), along with the appropriate body positioning and anatomical references. The selected measurement locations were based on Nagel et al.’s protocol for predicting body composition in preterm infants through ultrasound measurements of muscle and adipose tissue thickness, as well as the practical feasibility of measuring these areas in a supine position [12]. To prepare for image collection, the infant will be placed in supine position with all but the area of interest swaddled to maintain comfort and warmth.

Fig. 1.

A Biceps (biceps brachii and brachialis). Body position: supine with arm extended (palm up). Measurement point: ½ distance between acromion (shoulder) to antecubital crease of the arm on anterior centerline of the arm (shoulder and elbow). B Abdomen (rectus abdominis). Body position: supine. Measurement point: between costal margin and anterior superior iliac crest (bottom of rib cage to top of hip bone), to the appropriate side of the umbilicus (depending on side of infant used). C Quadriceps (vastus intermedius and rectus femoris). Body position: supine with knee extended and quadriceps relaxed. Measurement point: ½ distance from anterior superior iliac spine to the superior patellar border (hip to knee)

Three still images will be captured, followed by three video sweeps of each designated region. Example images are shown in Fig. 2. The video “sweep” data will generate a robust dataset rich in anatomical information for analysis. Unlike traditional methods that rely on precise operator-dependent image “freezing”, video sweeps provide continuous spatial sampling. This allows representative frames to be selected algorithmically, reducing the burden on the clinician and mitigating operator-dependent variability in busy or low-resource settings. Video sweep data will be used as inputs for AI-based analysis alongside still images, and feasibility outcomes will include evaluation of the completeness and quality of both acquisition types.

Fig. 2.

A Sample image of arm (biceps brachii and brachialis). B Sample image of abdomen (rectus abdominis). C Sample image of leg (quadriceps). SCF, subcutaneous fat; Bi, biceps brachii; Br, brachialis; RA, rectus abdominis; RF, rectus femoris; VI, vastus intermedius

For this study, a primary operator will perform the triplicate still images and video sweeps at all three anatomic sites. To evaluate inter-rater reliability, a second trained operator will independently record triplicate data at one randomly selected site (arm, abdomen, or leg). All clinicians and research staff involved in image acquisition will undergo standardized training on probe positioning, anatomical landmark identification, and minimal compression techniques. Competency will be established through supervised practice sessions and periodic image quality review to ensure consistency across operators and study sites. Finally, a radiologist associated with the study will review the collected images and videos to rule out any clinically significant findings that would affect the participant’s medical plan of care.

ML-based image analysis

Our prior work demonstrated the feasibility of predicting neonatal whole-body composition using deep learning models trained on static ultrasound images performed by a single operator in a single high-resource clinical setting [15, 16]. In contrast, the present study focuses on establishing the feasibility of a standardized multisite data acquisition protocol using handheld ultrasound devices and incorporating video sweep data to support scalable model development. For the exploratory analysis of collected images, we will employ deep learning models, including Unet, AttentionUNet, EfficientNet, and ResNet, to predict whole-body composition (FM and FFM) from regional ultrasound images and videos [26–29].

These models are well-suited for image segmentation and classification tasks. UNet and AttentionUNet utilize symmetric downsampling (encoding) and upsampling (decoding) layers, systematically reducing the size of the ultrasound image to capture essential features during encoding and then restoring it to its original resolution during decoding. The AttentionUNet enhances this process with attention mechanisms, allowing the model to selectively focus on the most important features within the image. Meanwhile, ResNet employs residual connections to facilitate the training of deeper models, and EfficientNet optimizes performance through scaling, achieving high accuracy with fewer parameters.

These models will be fine-tuned using the pilot data to optimize their performance and adapt to the specific characteristics of our clinical dataset. During training, the models will learn to extract relevant features from various scales within the images to provide precise predictions. The dataset will be divided into training, validation, and testing subsets. To ensure that the models generate clinically relevant predictions, we will use Gradient-weighted Class Activation Mapping (Grad-CAM), which produces heatmaps overlaying the ultrasound images [30].

This visualization will allow for clinical validation, highlighting specific anatomical areas of focus—such as the subcutaneous fat layer and musculoskeletal fascia—to ensure the models are prioritizing clinically relevant features rather than image artifacts. We will select the optimal deep learning model based on a combined analysis of their Grad-CAM heatmaps and their accuracy in predicting FM and FFM. The accuracy of these predictions compared to the ground truth (ADP) will be assessed using RMSE, MAPE, MAE, and MSE.

Discussion

Measuring human body composition, including FM and FFM, is essential for assessing health and nutritional status, understanding the impact of disease, and evaluating changes resulting from nutritional, therapeutic, or behavioral interventions. This is particularly true for infants and young children, because early-life growth patterns and nutrition are increasingly recognized to program later-life health outcomes [30, 31]. However, obtaining body composition measurements in infants can be challenging and often requires expensive equipment, specialized facilities, and trained personnel. With the advent of low-cost mobile ultrasound systems and their demonstrated effectiveness in assessing adipose tissue and skeletal muscle, this pilot study aims to test the feasibility and acceptability of ultrasound technology to acquire regional images and video for predicting whole-body composition in full-term and preterm infants [9, 10]. The results of this pilot study will provide important data to guide the design of a larger future study to develop a novel ultrasound tool that guides users in collecting high-quality data and automatically determining body composition metrics for nutritional evaluation.

This project has significant potential benefits for both society and the medical field, as it seeks to generate pivotal data that can create a more efficient means of measuring body composition, thereby facilitating improved assessment of growth and development. The ultimate goal is to develop a tool that enables frontline healthcare workers to measure body composition cost-effectively and without the need for specialized training to operate the device. Increasing access to body composition measurement tools can help mitigate challenges related to diagnosing and treating undernutrition and overnutrition, particularly in low-resource settings or within populations that require continuous medical care like hospitalized preterm infants. There is currently no standardized ultrasound protocol for measuring body composition, and, to the authors’ knowledge, no prior studies have utilized portable ultrasound systems for this purpose.

By implementing this pilot study in varied environments, across low- and high-resource settings as well as measuring both full-term and preterm infants, we hope to explore the device's applicability across various clinical contexts and demographics. This approach aims to enhance our understanding of the generalizability and effectiveness of ultrasound in assessing body composition. Ultimately, the findings from this study may contribute to improving the diagnosis and management of nutritional health issues, fostering insights into the relationship between nutritional health, nutrition-sensitive diseases, and overall quality of life.

Abbreviations

AI

Artificial intelligence

BMI

Body mass index

CNN

Convolutional neural network

FM

Fat mass

FFM

Fat-free mass

IRB

Institutional review board

OPD

Outpatient pediatric department

MAE

Mean absolute error

MAPE

Mean absolute percentage error

ML

Machine learning

MSE

Mean squared error

NICU

Neonatal intensive care unit

RMSE

Root mean squared error

Authors’ contributions

BR: conception of the work, design of the work, drafting the work, revising the work. MA: conception of the work, design of the work, drafted the work. JK: conception of the work, design of the work, and drafted the work. HC: design of the work, drafted the work. JH: design of the work, drafted the work. KH: design of the work, drafted the work. DW: conception of the work, design of the work, and revised the work. MB: conception of the work, design of the work, revised the work. TG: conception of the work, design of the work, revised the work. JE: conception of the work, design of the work, and revised the work. KB: conception of the work, design of the work, and revised the work. JP: conception of the work, design of the work, draft of the work, revise of the work. All authors read and approved the final manuscript.

Funding

Funding for this study was provided by the Google Research Scholar Program, Google Award for Inclusion Research, the Boston College Schiller Institute for Integrated Science and Society, and the Boston College Undergraduate Research Fellowship Program. KB is supported by the National Institute of Child Health and Human Development (K23HD104000).

Data availability

Since this manuscript describes a pilot protocol, data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

Declarations

Ethics approval and consent to participate

Ethical approval has been obtained from the Institutional Review Boards (IRBs) of Boston College, Mass General Brigham (protocol #2023P000325), and Jimma University, as well as the Ethiopian Ministry of Education National Research Ethics Review Committee (protocol #17/242/731/23) to safeguard participant rights throughout the study. Written informed consent will be obtained from the parents or legal guardians of all participants prior to enrollment in the study. All data will be collected and handled in accordance with ethical guidelines to ensure participant confidentiality throughout the study.

Consent for publication

Not applicable. The manuscript does not contain individual person’s data in any form, including details, images or videos.

Competing interests

The authors declare that they have no competing interests.