A Novel Adaptive Swallow Swarm Optimisation Method for Medical Image Segmentation in Brain MRI Using Machine Learning

Abstract

Automatic segmentation has to be effective to provide sufficient results for clinical use due to inherent challenges involved in medical images, including poor image quality, different imaging, segmentation protocols or variations in patients. Further, they are reduced to previously invisible classes by a lack of image-specific adaptation and the absence of generality. When handling some complex cases, it may lead to ineffectiveness, taking into account the differences in distribution between training and test data. To solve these issues in the proposed work, noise removal and segmentation algorithm is introduced by using Adaptive Swallow Swarm Optimization (ASSO) with Convolutional Neural Network (CNN) and is called as ASSO-CNNseg to find traces of cancer in Brain Tumour images. In the noise removal step nonlinear filtering algorithm is introduced to remove noises from the image samples efficiently. After the removal of noises from the image samples, then segmentation is performed by using the deep learning algorithm consists of some steps: image-based fine tuning, interaction-oriented uncertainty, and segmentation process. In the fine-tuning algorithm, ASSO algorithm is introduced for weighted loss feature taking the network and interaction-based uncertainty taken into account during image-dependent fine-tuning. Next, here the initial effort make use of CNNseg to tackle their earlier unknown objects are the image segmentation context is presented. The experimental outcomes demonstrate the highly accurate results of the proposed ASSO-CNNseg method with less number interactions between users and less consumption of time for user compared to classical interactive segmentation techniques such as Slic-seg and BIFSeg.