ARCH Stakeholder Collaboration Projects (SCP)

Adolescent Early Diabetes Diagnosis in American Samoa through AI Image Analysis of Acanthosis nigricans

ARCH Project Type:  AI enrichment of community based research

Health Equity Focus: NCD in indigenous Pacific Island populations

Principal Investigator: Mata’uitafa Solomana Faiai

Affiliations:  PhD Student in Chronic Disease Epidemiology, Yale School of Public Health

Center for Indigenous Innovation and Health Equity Fellow, Chaminade University

NSF INCLUDES ALL-SPICE Data Science Alliance Researcher in Residence,

ARCH AI Navigators: Stokes, Burkhardt, Saini

Project Background:

High rates of cardiometabolic disease in American Samoa are significant health burdens. CIIHE Fellow Mata`u Faiai is examining drivers of adolescent cardiometabolic settings in cardiometabolic health in American Samoa, reasoning that there are opportunities to intervene and mitigate disease risk in this under-studied group. Faiai’s central research goal is to the central premise of this thesis is to understand the development and distribution of obesity, diabetes and hypertension and its key drivers in the adolescent population in American Samoa. Using a socio-ecological model  (SEM), she is working to Identify factors impacting healthful nutrition among American Samoa adolescents (SEM: relationships and community), Identify factors influencing obesity, diabetes, and hypertension in adolescents of American Samoa (SEM: Individuals) and examine the influences of  friendship networks on adolescent obesity, diabetes, hypertension in American Samoa (SEM: Community and Peers). central to this work is a school based study that engages >1000 American Samoan adolescents in diet, physical activity, depression and body image screening tools that have been validated among adolescents and Samoan adults, clinical markers (HbA1c, BMI) and photographic morphometry as well as social network analyses using an exponential random graph model (ERGM). The proposed ARCH Collaboration leverages this study and explores an exciting new potential for early diagnosis T2D in the adolescent population using the application of AI guided image analysis to morphometric participant photographs. Acanthosis nigricans (AN) is a skin condition characterized by brown-to-black, poorly defined, velvety skin hyperpigmentation in body folds and creases. It typically occurs in individuals younger than age 40, is associated with insulin resistance, T2D,  obesity or endocrinopathies. The current measurement of AN for both children and adults is subjective. A single trained observer can score a child with AN on a scale of 0 to 4. A score of 0 is defined as not detectable on close inspection, a score of 1 to 4 are defined as present, mild, moderate or severe respectively.

Research Question: Is physical appearance of Acanthosis nigricans a marker for early detection of T2D-related cardiometabolic changes, mitigating under-diagnosis and accelerating intervention in adolescent American Samoans?

Hypothesis: AI guided image analysis will detect skin changes in Acanthosis nigricans in adolescent American Samoans as an early marker of diabetes.

Proposed AI Methods:

• collecting ethnically-appropriate diverse set of annotated medical images that indicate the presence or absence of Acanthosis nigricans, standardize images for size and quality, and augment the dataset with variations like rotation and color adjustments to enhance model robustness;

• test approach such as a convolutional neural network (CNN) architecture, explore leveraging a pre-trained model to expedite training and enhance feature recognition

• train the model and optimize performance 

• assess effectiveness with metrics such as accuracy and AUC-ROC

• validate the model's predictions with dermatology experts to ensure clinical reliability