Brain MRI Tumour Localization and Segmentation Through Deep Learning

The segmentation of magnetic resonance imaging (MRI) is an important task in medical imaging, particularly for brain MRIs, where accurate segmentation of anatomical structures, often challenged by uneven shapes and fuzzy boundaries of tumours, is difficult to achieve. Thus, automating reliable segmentation of the region of interest (ROI) is essential in medical imaging. This work introduces an automated approach for brain tumour segmentation, leveraging a Deep Convolutional Neural Network (CNN) with a focus on effectively segmenting brain tumours within T1-weighted contrast-enhanced MRI (CE-MRI) images. The proposed method is first performed on the classified images to localize the tumour regions of interest(ROIs). In the next stage, the algorithm contours the concentrated tumour boundary for the segmentation process, which contains the network and attention module, both spatial and channel. To evaluate the overall system's performance, precision, recall, Jaccard index, and dice similarity coefficient (DSC) were calculated, where we achieved 0.8757, 0.8804, 0.8624, and 0.8995, respectively. Our approach demonstrates promising results compared to previous methods using the same database.