A Study on Disease Classification Methods for Ophthalmic Ultrasound Images

Ocular diseases represent a significant public health concern in China. Given that early detection and timely intervention are critical to preventing irreversible vision loss, the development of efficient and accurate diagnostic approaches is of paramount importance. This paper proposes an intelligent classification method for ophthalmic diseases based on an improved EfficientNetV2 model, designed to distinguish three common and easily confounded conditions. To address the challenge of similar visual features among certain eye diseases in ultrasound images, the YOLOv11s model is employed to accurately detect and localize the eyeball region in color ultrasound images. A standardized dataset is then constructed based on the detected regions. Considering the long-tail distribution characteristic of the dataset, the EfficientNetV2 architecture is enhanced and trained for classification. Experimental results demonstrate that the proposed method achieves an accuracy of 90.49% in the classification task, highlighting its potential for assisting in the intelligent diagnosis of ocular diseases, particularly in areas with limited medical resources.