ADVANCED DEEP LEARNING TECHNIQUES FOR BRAIN TUMOR SEGMENTATION AND CLASSIFICATION
TL;DR
The experimental results confirm that the proposed method can achieve high performance, accuracy, and reliable results, and has the potential capacity to assist doctors in distinguishing between benign and malignant tumours, helping them make the best decision to save victims of brain cancer.
Abstract
Problems with uncontrolled cell growth in the human body are a major cause of cancer, like brain cancer, which occurs in the brain or the nervous system. Tumours are categorised as benign or malignant and are defined by cellular activity and structure. Cancerous tumours are invasive and might require surgical removal, so early and accurate diagnosis is crucial. A radiologist-based diagnosis manually is time-consuming and error-prone, especially in denseness regions. This work aims to solve this problem through the design and application of computer-aided design (CAD) approaches for tumour detection and classification of MRI data in the brain. The method starts from the pre-processing of images by Anisotropic Diffusion with Median Filtering (ADWMF) for noise removal and for the enhancement of the tumour edge. For the segmentation, we employ the IHMFKC algorithm. Feature extraction is performed based on GLCM and FOS. In total, five classifiers (i.e., YOLOv3, Faster R-CNN, ResNet-50, PNNsf3, KNN) are realised via IHMFKC. As a result, IHMFKC embedded with PNN and YOLOv3 obtained classification accuracies of 96.1% and 95.71%, respectively. The experimental results confirm that the proposed method can achieve high performance, accuracy, and reliable results, and has the potential capacity to assist doctors in distinguishing between benign and malignant tumours, helping them make the best decision to save victims of brain cancer.