Optimized DNetCNN Framework for Brain Tumour Segmentation and Classification Using MRI Images

Abstract

The primary cause of brain illnesses is irregular brain cell proliferation, which can harm the structure of the brain and ultimately result in malignant brain cancer. Using a computer-aided diagnostic system to enable an early diagnosis and immediate therapy involves dealing with difficulties, foremost among which is the precise identification of various diseases using magnetic resonance imaging (MRI) scans. This research proposes a new Darknet Convolutional Neural Network (DNetCNN) framework for accurate diagnosis of pituitary, meningioma, and glioma, as well as a two-step preprocessing method to improve the quality of MRI images. The relevant features are extracted using the ResNet152V2 technique to accurate the ground-truth segmentation of tumours. Finally, to improve the classification accuracy using the Enhanced Rat Swarm Optimization Algorithm (ERSOA). An analytical comparison is made between the proposed framework and other models covered in this study. Tested on a dataset comprising MRI pictures, an exceptional competitive accuracy of 99.78% is obtained, with 99.78% accuracy in recognizing gliomas, 99.86% accuracy in detecting meningiomas, 99.68% accuracy in detecting pituitary tumours, and 99.80% accuracy in identifying normal images. The proposed architecture's resilience is demonstrated by experimental results, which have also quickly improved the accuracy of brain disease diagnosis.