ABSTRACT
Type 2 diabetes mellitus is one of the most common metabolic diseases in the world. However, frequent blood glucose
testing causes continual harm to diabetics, which cannot meet the needs of early diagnosis and long-term tracking of
diabetes. Thus non-invasive adjuvant diagnosis methods are urgently needed, enabling early screening of the population
for diabetes, the evaluation of diabetes risk, and assessment of therapeutic effects. The human eye plays an important
role in painless and non-invasive approaches, because it is considered an internal organ but can be easily be externally
observed. We developed an AI model to predict the probability of diabetes from scleral images taken by a specially
developed instrument, which could conveniently and quickly collect complete scleral images in four directions and
perform artificial intelligence (AI) analysis in 3 min without any reagent consumption or the need for a laboratory. The
novel optical instrument could adaptively eliminate reflections and collected shadow-free scleral images. 177 subjects
were recruited to participate in this experiment, including 127 benign subjects and 50 malignant subjects. The blood
sample and sclera images from each subject was obtained. The scleral image classification model achieved a mean AUC
over 0.85, which indicates great potential for early screening of practical diabetes during periodic physical checkups or
daily family health monitoring. With this AI scleral features imaging and analysis method, diabetic patients’ health
conditions can be rapidly, noninvasively, and accurately analyzed, which offers a platform for noninvasive forecasting,
early diagnosis, and long-term monitoring for diabetes and its complications.
CONCLUSIONS
In summary, we have developed a non-invasive AI method to predict the risk of type 2 diabetes: We have developed a
MIL model that predicts the probability of lung cancer through scleral images taken by a specially developed instrument,
which can conveniently, quickly acquire complete scleral images in four directions, complete AI analysis within 3
minutes, without any reagent consumption, and no laboratory. The average AUC of the binary classification results of
the MIL model is 0.876, indicating that there is great potential for early screening of type 2 diabetes in regular physical
examinations or daily family health monitoring.
Our results suggest a new concept that in this innovative study, the use of deep learning to analyze scleral images can
help detect type 2 diabetes. This work supports a potential step towards the development of deep learning-based tools for
pre-screening diabetes probability assessment in outpatient clinics or diabetes screening in the community, which may
help guide further diagnostic tests or visits.
Authors:
Lv, Wenqi, Fu, Rongxin, Lin, Xue, Su, Ya, Jin, Xiangyu, et al.
SPIE 11900, Optics in Health Care and Biomedical Optics XI, 1190013 (15, October 2021); doi: 10.1117/12.2601222
Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing 100084, P. R. China
Full Abstract: https://iridology-research.com/pdf/1190013.pdf
