Development of Brain MRI Image Segmentation Methods Based on Computer Vision and Deep Learning

Brain Magnetic Resonance Imaging (MRI) is routinely utilized for observing cerebral diseases. During the diagnostic process, specialists are tasked with measuring the scope of brain lesions. However, manual measurements can be both time-consuming and imprecise. Computer vision presents a solution capable of achieving automated measurements. brain MRI images comprise the skull and brain tissues. Prior to analyzing the brain tissues, it is imperative to first separate the skull images, which then paves the way for subsequent analysis of brain tissue images. Accordingly, this study develops an image processing method for skull image separation. The method aims to retain images of both the cerebrum and cerebellum and segregates the brain MRI images into three sections based on two features. The Convolutional Neural Network YOLOv7 is employed for recognizing these two features. Specific techniques are devised to separate skull images corresponding to the two sections (containing cerebrum and cerebellum tissues) to retain the essential parts of brain images. To boost recognition accuracy, sample image training focuses on the Region of Interest (the temporal lobe and ocular tissues). Such an approach results in high recognition accuracy (average loss=0.015, mAP=0.99, Precision=0.95, Recall=0.95). Moreover, the study conducts numerous experiments to validate the method's efficacy.

I. Introduction

In the past decade, the field of smart healthcare has garnered extensive attention and recognition, primarily due to the rapid advancements in medical strategies, healthcare diagnostic equipment, and medical data analysis driven by artificial intelligence. These developments have significantly enhanced the accuracy of medical diagnostics, enabling healthcare professionals to make more precise diagnoses, treatments, and caregiving, while also drastically reducing the risks of medical errors. Smart healthcare technology not only aids healthcare practitioners in improving disease management but also elevates the overall quality of medical services within hospitals, thereby enhancing the overall standard of healthcare delivery. While current artificial intelligence-based computer image detection systems cannot replace the role of physicians in diagnosis, they undeniably serve as effective tools to assist physicians in detecting subtle lesion details that might be difficult to discern, as well as in processing vast volumes of continuous images to extract relevant feature values and statistical data. These image detection systems not only save valuable time for physicians but also allow them to observe lesion development trends through data, enabling more accurate diagnoses. Furthermore, these systems aid physicians in swiftly and accurately identifying lesions, understanding disease progression, and formulating more efficient treatment plans.