Beyond MobileNet: An Improved MobileNet for Retinal Diseases

Wenhui Zhu, Peijie Qiu, Xiwen Chen, Huayu Li, Hao Wang, Natasha Leporé, Oana M. Dumitrascu, Yalin Wang

Abstract: Myopic Maculopathy (MM) is the leading cause of severe vision loss or blindness. Deep learning-based automated tools are indispensable in assisting clinicians in diagnosing and monitoring RD in modern medicine. Recently, an increasing number of works in this field have taken advantage of Vision Transformer to achieve state-of-the-art performance with more parameters and higher model complexity compared to Convolutional Neural Networks (CNNs). Such sophisticated model designs, however, are prone to be overfitting and hinder their advantages in specific tasks in medical image analysis. In this work, we argue that a well-calibrated CNN model may mitigate these problems. To this end, we empirically investigated the macro and micro designs of a CNN and its training strategies by starting with a standard MobileNet. Based on the investigation, we proposed a lightweight MobileNet training framework equipped with a series of optimal parameters and modules based on retinal images. As a result of performance, our model secured third place in the MICCAI MMAC 2023 Challenge - Classification of Myopic Maculopathy. Our software package is available at https://github.com/Retinal-Research/NN-MOBILENET