A Novel Approach in Bio-Medical Image Segmentation for Analyzing Brain Cancer Images with U-NET Semantic Segmentation and TPLD Models Using SVM

Abstract

Many medical applications need to be able to separate and find brain tumor's using CT scan images. There have been a lot of recent studies that used distinguish between benign and malignant tumour to find out where and how big a tumour is. Even though they did well at segmenting the Medical Image Segmentation Decathlon (MISD) dataset, their complex structure requires more time for training and analysis. To build a flexible and efficient brain tumour segmentation system, we offer a pre-processing method that only works on a small part of the images instead of the whole Image. U-Net with three parameters Deep Learning models can be trained more quickly and with less overfitting with this method. Support vector machine is used in the second stage because there are fewer brain images for each slice. When U-Net+SVM looks at data this way, it can find both local and global features in it. The Three parameter method had shown to be more accurate at separating brain tumors from healthy parts of the brain than other models. The U-Net+SVM+Three Parameter Features method requires the tumour to be in the middle of the model and to be there. A lot of testing on the Medical Image Segmentation Decathlon (MISD) dataset showed that our model can get good results: Dice scores for overall cancer, more cancer and the core of the tumour are all 96%, which is the same for all three.