Research on Fusion of Clinical Indicators and Pathological Imaging for Diabetic Retinopathy Classification and Early Diagnosis

Meaning Diabetic retinopathy (DR) is a leading cause of blindness in diabetic patients with early diagnosis is crucial for preserving vision. It is a proposal of classification of the diabetes retinopathy by early diagnosis based on the fusions of the clinic indicators and pathological images. For the feature extraction from retinal images, we implemented a single Convolutional Neural Network (CNN). In this model, we leverage clinical data as ancillary information for images to enhance the diagnostic ability of the model via fusing with image features. To be more specific, the neural network structure has been modified to handle the temporal information in the clinical data, and weighted and coupled with the image features obtained by CNN. To enhance its performance, we incorporated a self-attention mechanism to refine the feature fusion of images and clinical data and utilized a hybrid optimization strategy to train the model. The experimental results demonstrate that the model is effective in classifying DR and performing early diagnosis, with accuracy up to 92% and sensitivity for early diagnosis to 90% which is much more than the traditional method, indicating that the method has great clinical application value.