A Deep Learning Approach for Brain Tumor Firmness Detection Using YOLOv4

Diagnosing brain tumors typically starts with an investigation of the brain's Magnetic Resonance (MR) imaging. However, such a manual analysis of MR images is tedious and time consuming. Moreover, it may be inaccurate and subject to the radiologist's opinion. Diverse machine learning models have already been built to map the visual properties of brain magnetic resonance images (MRIs) into predefined tumor classes such as “malignant” and “benign”. Further, the discrimination between firm and soft meningioma determines the appropriate patient counseling as well as the type of operative procedure to be adopted to cure this primary tumor of the brain. In this research, we report on a deep learning approach for the automatic firmness detection of meningioma brain tumors using MRIs that deploys a YOLOv4 model to learn the association between the low-level features of these images and the firm/soft classes. Our testing results, obtained using a real collection of meningioma MRIs from a local university hospital, show that the employed approach based on YOLOv4 outperforms state-of-the-art relevant solutions with specificity, F1-score, and sensitivity values of 95.83%, 99.03%, and 100%, respectively.