Brain tumor detection using unsupervised local center of mass segmentation: As an CNN based approach

Segmentation of Brain tumor is a very important as well as difficult issue in medical image processing field, as human aided manual grouping might end in improper prediction and analysis. Moreover, it’s a difficult work when a huge volume of information to be processed is present. Because brain tumors possess a variety of share and appearances similarities with extracting tumor, normal tissues areas of pictures become hard. In general, medical images are accomplished with significant details associated to organs, bones and tissues. Hence, it is essential to select suited technique of image processing for extracting clinical data without any compromising on feature interest of original region. However, the conventional approaches such as CT, X-RAY and MRI have generated various images of musculoskeletal, joints and bones regions with several degree of accuracy. As a result, image segmentation is an important part of the medical imaging process. Thus, the automated segmentation technique has demonstrated enhanced utilisation by reducing human mistakes in medical image processing, and it has improved illness detection or screening without any faults. This paper purposes to generate an unsupervised segmentation approach for properly analyzing the categorization of brain MRIs with reduced computing time and improved classification accuracy. This work offers a new novel system for unsupervised image segmentation centered on the calculation of LCM, Local Center of Mass. Finally, the tumor area is segmented and categorized using a Convolution Neural Network (CNN) to find its kind. The suggested CNN, which is constructed using Keras and Tensorflow, outperforms classic CNNs. CNN achieved an accuracy of 95.65 percent in our work, which is really impressive.