A Deep Learning Algorithm Improvement for Brain Tumour Segmentation Using Kernel-Based CNN and M-SVM

Recently, medicinal imaging has shown significant growth in demand due to its ability to provide useful information from human anatomical images, which helps understand symptoms and diagnose patients (Garzón & Cardenas-Villa 2020), with astrocytoma’s separation among the crucial tasks popular the field that can assist clinicians on diagnosis, treatment planning as well as monitoring of patient they are concerned about. Deep learning algorithms have started to show impressive results in the segmentation of brain tumors from MRI images over recent years. Therefore, this study presents a new deep learning algorithm that utilizes kernel-based Convolutional Neural Networks (CNN) and multi-class Support Vector Machines (M-SVM) for more correct separation of astrocytoma’s. The latter is a kernel-based CNN method that efficiently learns deep structures as of the contribution MRI pictures. The features remain input into the mm classifier, which accurately classifies them to different tumor regions (necrotic core, edema and enhancing tumor). Besides, we introduce a boundary refinement module into the framework to fine-tune results at segmentation boundaries for better accuracy. Here, we attempted to analyze our analyze algorithm over publicly available annotation datasets experimentally and compared them against the state-of-the-art as baselines. Our approach outperformed previous studies with an normal Dice comparation constant of 0.93, illustrating the capacity to segment brain tumor regions accurately. In addition, our algorithm demonstrated a resilience to noise and differences in image resolution.