Deep Learning And The U-Net Model For Magnetic Resonance Image Segmentation Of Brain Tumors

Abstract

Due to the general high mortality rate associated with brain cancer, it is crucial to diagnosis them early so that they can be treated and mortality reduced. Unwanted cell multiplication in the human brain causes brain tumors, which are deadly and life-threatening diseases. Brain tumors are one of the most difficult disorders to treat among the many diseases found in medical science. Tumor are classified as benign or malignant, with benign tumor being non-cancerous and malignant cancers being cancerous. In this paper, the proposed method consists of three basic steps: tumor image preprocessing, segmentation and classifications. As the first stage of preprocessing, noise removal is performed. To remove the noise component and improve the image intensity, the adaptive median filter technique is applied. The segmentation stage comes next. U-net is among the most significant semantic segmentation framework for a convolutional neural network. In the biomedical imaging sector, it is commonly utilised for lesion segmentation, anatomic segmentation, and classifications. Lastly, the result is classifications using convolutional neural network (CNN) classifiers. The precision, sensitivity, specificity, accuracy, dice, jaccard are calculated from the developed model. The experimental result has proved that CNN model is performing with accuracy 96 percent.