EvoU-Net
Tahereh Hassanzadeh, Daryl Essam, Ruhul Sarker
Abstract
Developing a Deep Convolutional Neural Network (DCNN) for image segmentation is challenging research topic which needs extensive experiments to find an appropriate network structure and a precise set of hyper-parameters. The limited number of available labelled images and the required computational infrastructure make this task even more challenging. Evolutionary strategy is an optimisation technique that is applicable to alleviate the above difficulties. This paper proposes an evolutionary-based method to find a precise and small network for medical image segmentation. To the best of our knowledge, EvoU-Net is the first evolutionary method to develop an U-Net-based deep network topology with for medical image segmentation. In the proposed model, a Genetic Algorithm (GA) is applied to design an optimal network structure, along with its parameters, for MRI image segmentation. EvoU-Net outperformed U-Net and AdaResU-Net while using less than 10% and 50% of trainable parameters respectively, for segmentation of a publicly available prostate MRI dataset.