Brain Tumour Detection Through Modified UNet-Based Semantic Segmentation

2. Related Work

Image segmentation is an important part of many visual comprehension systems. It entails dividing an image (or a video frame) into several segments or objects (Naidu et al., 2018). Medical image analysis (e.g., tumour border extraction and organ volume assessment), autonomous vehicles (e.g., accessible surface and pedestrian recognition), surveillance cameras, and virtual reality are just a few of the applications that use segmentation (Forsyth & Ponce, 2011). Whereas, semantic segmentation classify each pixel into respective labels or masks, means it assigns a particular class to all pixels in an image. Deep learning approaches have recently become the illegitimate norm for a variety of medical image processing applications. To segment any image, the existing techniques (Liu et al., 2021) rely on understanding of digital image analysis and mathematics. The computation is simple, and the segmentation generally quick, but the correctness of the segmented not guarantees the carry of much details about the image (Berinde & Ţicală, 2021). By autonomously learning a structure of image features, unsupervised learning algorithms avoid the difficulty of defining and selecting features (Dong et al., 2020). Random forests (RF) and support vector machines (SVM) are perhaps the most effective supervised learning algorithms with discriminative classifiers for effective brain tumour identification (Dabija et al., 2021).