Brain MRI Image Classification for Tumor Detection Using Integrated Hybrid Convolutional K-Nearest Neighbor Model

In the field of medical image processing, brain tumor segmentation is one of the most important and challenging jobs since manually categorization by humans can lead to incorrect diagnosis and prognosis. Furthermore, it is a frustrating chore when there is a lot of data that has to be helped. Because brain tumors have a wide range of appearances and normal tissues and tumors are similar, it is difficult to separate specific tumor areas from pictures. Keeping this in mind, a preliminary processing method for brain MRI is presented in this study that applies Otsu's Thresholding and Morphological operation. An up-to-date online image dataset (consists of 3064 slices of brain images containing samples of meningioma, glioma, and pituitary tumor types) from 233 patients with a variety of tumor sizes, positions, forms, and intensity values of images is used for the experimental investigation. Lastly, we used Convolutional Neural Networks (CNN) and K-Nearest Neighbors (KNN) in the classical classification section. The hybrid Convolutional K-Nearest Neighbors (CKNN) model was then used, which produces superior results than the conventional used models. The primary goal of this study was to use brain MRI images to identify brain tumors. This study showed significant performance with accuracy of 89.88% for the hybrid CKNN model.