Retina Vessel Segmentation Using Lightweight U-net Like Model Optimized for Neuromorphic Crossbar Array

Segmentation of retinal vessels in fundus images is a key computational task for assessing retinal health. Deep learning (DL) methods, particularly convolutional neural networks (CNNs), are commonly used for automating medical image segmentation and disease classification. However, there is growing interest in developing lightweight, hardware-efficient deep learning models that are energy-efficient and suitable for real-time data processing, especially in image processing tasks. Neuromorphic computing offers a means to execute artificial neural networks quickly and with reduced energy consumption. By considering a software-hardware co-design, we proposed a U-net like CNN model that is lightweight and optimised for neuromorphic hardware for retinal vessel segmentation in fundus images. This model has the potential to facilitate real-time processing of fundus images in handheld, portable eye cameras, enhancing user experience. The proposed model demonstrated image segmentation performance comparable to other commonly used complex CNN architectures for image segmentation tasks.