Research on the relationship between intelligent diagnosis of diabetes retinopathy and biomarkers based on multimodal data fusion

BACKGROUND AND PURPOSE: To explore the risk factors of diabetic retinopathy (DR) in type 2 diabetes (T2DM) patients, construct a multimodal deep neural network model based on clinical risk score and fundus image characteristics, and verify the expression changes of key proteins in DR rat models, so as to provide a new method for early diagnosis and grading evaluation of DR. METHOD: < 0.05). RESULT: The multimodal deep neural network model that integrates clinical risk scores with fundus imaging features can significantly improve the accuracy and discrimination of DR grading diagnosis, and has good generalisation ability. CONCLUSION: The DR rat model exhibits significant inflammatory response, oxidative stress, and abnormal angiogenesis, with upregulation of SPP1, TMEM173, and Circp protein expression, which may be involved in the pathogenesis of DR.