Advanced Feature Rich Medical Image Segmentation Based on Deep Feature Learning

Deep learning approaches have demonstrated significant potential in automating this activity, hence removing the necessity for radiologists to manually annotate images. Concatenate blocks are incorporated in the suggested model in order to learn various filters or features that are utilised to extract relevant information from the image. Since k-fold validation is a strong and adaptable method for assessing deep learning model performance, it is used to train the suggested model. By setting the value of k to 5, the suggested model is tested on five distinct data subsets in order to produce the optimised model and produce more accurate results. Ten percent was used for validation, twenty percent for testing, and seventy percent for training. The suggested model was built in four stages: (a) picture classification, (b) feature attribute selection, (c) tumour segmentation, and (d) noise removal. We suggest using deep neural networks, trained on MRI scans of patients’ brains, to detect tumours. For tumour segmentation, the suggested architectures rely on inception modules and convolutional neural networks.