The Use of Deep Learning for Brain Tumor Classification: A Study of Cropped, Uncropped and Segmented Lesion Images of Varying Size

Abstract

Deep Learning has recently attracted a lot of interest from academics as the most cutting-edge trend in the machine learning
area. Deep learning has shown to be an effective machine learning method, and it has found several uses in tackling complicated
issues that call for a high level of sensitivity and precision, such as those in the medical industry. One of the most prevalent and
deadly malignant tumor disorders, brain tumors often have an extremely low life expectancy when detected at advanced stages.
Thus, following tumor detection, grading is an essential step in developing an efficient treatment strategy for brain tumors. The
authors of this study graded 3064 T1 weighted contrast-enhanced brain MR images for tumors into three categories: gliomas,
meningiomas, and pituitary tumors. CNN is a popular deep learning architecture that they employed for this task. With an overall
performance of 98.93% accuracy and 98.18% sensitivity for the cropped lesions, the proposed CNN classifier is a powerful
tool. When applied to uncropped lesions, the results are 99% accuracy and 98.52% sensitivity. When applied to segmented lesion
images, the results are 97.62% accuracy and 97.40% sensitivity.