Improvement of an eye disease detection model by using the denoising diffusion implicit model

Abstract

With rapid developments in artificial intelligence (AI), the discussion about and applications of generative AI have increased substantially. Generative AI has extensive and valuable applications in many industrial and medical fields and is a possible solution for industries that struggle to collect large quantities of data. The present study evaluated the use of generative AI in eye disease prediction. Because retinal images are difficult to acquire, this study used a generative AI model [i.e., the denoising diffusion implicit model (DDIM)] to conduct data augmentation, thereby improving the accuracy of a convolutional neural network (CNN) model developed for eye disease detection. This study adopted the DDIM primarily for its high inference speed and ability to consistently generate high-quality samples in a limited number of steps, making it suitable for tasks that require high-quality medical images. With the increasing prevalence of electronic products, the number of patients with retinopathy or optic neuropathy is increasing annually, and patients are experiencing these diseases at increasingly younger ages. Moreover, eye diseases such as glaucoma and macular degeneration are becoming increasingly common in modern society. The developed CNN model exhibited a 3 % higher accuracy when it was trained using the data generated by the DDIM than when it was trained without these data. This CNN model can screen eye disease symptoms early to enable patients to receive timely treatment, thereby mitigating the risk and consequences of eye diseases. The results of this study indicate that the training data generated using the DDIM can enhance the accuracy of early eye disease detection.