Toward Biological Plausible Learning: A Novel Dendritic MnasNet Model

Abstract

Glaucoma is one of the leading causes of irreversible blindness worldwide. Deep learning methods have made significant strides in predicting glaucoma in recent years. However, existing models continue encountering challenges in extracting complex and subtle pathological features from fundus images associated with glaucoma. To address this limitation, we propose a novel DMNet model, which aims to enhance the integration of input signals by simulating the dendritic neuron model. This approach can improve the capture of fine details within glaucoma images and significantly boost classification performance. Experimental results indicate that DMNet outperforms traditional deep learning models on the glaucoma fundus image dataset, demonstrating its substantial performance advantages.