TY - GEN
T1 - DiabEye-Q
T2 - 2nd International Conference on Artificial Intelligence on Healthcare, AIiH 2025
AU - Khan, Sulaiman
AU - Biswas, Md Rafiul
AU - Shah, Zubair
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - Despite significant advances in diabetes classification, early prediction of diabetes onset using longitudinal multimodal data remains underexplored. In this study, we integrate clinical data with ophthalmoscopic images to predict diabetes onset among Qatari adults. his longitudinal study performs case—control analysis of 2,041 participants, including 1,076 males (246 cases, 830 controls) and 965 females (184 cases, 781 controls). We investigated the relationship between retinal features and diabetes status by extracting fractal geometry features and employing ANOVA for statistical validation. Furthermore, we develop a pipelined architecture that fuses XGBoost with a vision transformer (ViT) to identify risk factors associated with diabetes development. Age-stratified analysis reveals that while the model tends to overpredict diabetes in younger individuals (22–40 years), prediction accuracy improves markedly in older age groups (particularly 56–68 years). Additionally, gender analysis indicates a higher predisposition for diabetes among males compared to females. Our integrated model outperforms both standalone ViT and XGBoost, achieving 88.08% accuracy, 93% AUROC, and 88% recall. These findings underscore the potential of ophthalmoscopic imaging as a rapid, non-invasive screening tool for early diabetes detection.
AB - Despite significant advances in diabetes classification, early prediction of diabetes onset using longitudinal multimodal data remains underexplored. In this study, we integrate clinical data with ophthalmoscopic images to predict diabetes onset among Qatari adults. his longitudinal study performs case—control analysis of 2,041 participants, including 1,076 males (246 cases, 830 controls) and 965 females (184 cases, 781 controls). We investigated the relationship between retinal features and diabetes status by extracting fractal geometry features and employing ANOVA for statistical validation. Furthermore, we develop a pipelined architecture that fuses XGBoost with a vision transformer (ViT) to identify risk factors associated with diabetes development. Age-stratified analysis reveals that while the model tends to overpredict diabetes in younger individuals (22–40 years), prediction accuracy improves markedly in older age groups (particularly 56–68 years). Additionally, gender analysis indicates a higher predisposition for diabetes among males compared to females. Our integrated model outperforms both standalone ViT and XGBoost, achieving 88.08% accuracy, 93% AUROC, and 88% recall. These findings underscore the potential of ophthalmoscopic imaging as a rapid, non-invasive screening tool for early diabetes detection.
KW - Diabetes
KW - Longitudinal
KW - Ophthalmoscopic images
KW - Retinal
KW - Vision transformer
UR - https://www.scopus.com/pages/publications/105017225096
U2 - 10.1007/978-3-032-00652-3_21
DO - 10.1007/978-3-032-00652-3_21
M3 - Conference contribution
AN - SCOPUS:105017225096
SN - 9783032006516
VL - 16038
T3 - Lecture Notes In Computer Science
SP - 293
EP - 306
BT - Artificial Intelligence In Healthcare, Aiih 2025, Pt I
A2 - Cafolla, D
A2 - Rittman, T
A2 - Ni, H
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 8 September 2025 through 10 September 2025
ER -