Abstract: Deep learning has become increasingly prevalent in the medical field, with convolutional neural networks (CNNs) often used to analyze medical images. Among the various artificial intelligence technologies available, radiomics plays a crucial role in the detection and evaluation of glioblastoma from magnetic resonance imaging (MRI) scans, a challenging task for oncologists. In this study, deep learning techniques were applied to MRI brain scans obtained from publicly available datasets. Three CNN architectures were used: a sequential model for image classification, the U-Net architecture for image segmentation, and the You Only Look Once (YOLO) algorithm for object detection. Experiments were conducted to verify the effectiveness of the proposed approach and the performance was assessed using relevant parameters and metrics. The results obtained were then presented, analyzed and discussed to evaluate the performance of the models. The sequential model for image classification achieved a remarkable F1 score of 95%. The U-Net architecture for image segmentation initially underperformed with the default parameters but was able to achieve an impressive F1 score of 89% after adjusting the relevant parameters. However, the YOLO algorithm for object detection in images yielded a high rate of false-positive detections. Overall, the evaluated architectures demonstrated promising results in the detection and localization of glioblastoma tumours in MRI.

Keywords: deep learning, glioblastoma, convolutional neural networks, magnetic resonance images, image classification, image segmentation, object detection

Pages: 619-632