Integrating U-Net with Lightweight Cnns for Mri Brain Cancer Detection: a Comparison Among Mobilenetv2, Xception, and Alexnet

Brain cancer involves abnormal cell growth in the brain or nearby regions. MRI aids in segmenting tumor components like edema and core areas, though accuracy is challenging due to variations in tumor location, size, and shape. In a recent short paper, we applied an integrated CNN model that combines certain aspects of both ResNet50 and U-Net, leveraging the advantages and strengths of each. Specifically, we replaced the last 5 layers of ResNet50 with 10 additional layers, thereby serving as the encoder in the U-Net architecture. We employed the same scenario for lightweight architectures such as MobineNetV2, Xception, and AlexNet architectures. This study uses MR brain image datasets from Kaggle, paired with manually segmented FLAIR abnormality masks from TCIA for 110 patients (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{3, 9 2 9}$</tex> images) from TCGA's lower-grade glioma collection. Our results with ResNet50 show precision and IoU of 0.98, 0.97, respectively. Further results with Xception, MobineNetV2, and AlexNet show an accuracy of approximately <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.98,0.985$</tex>, and 0.99, respectively. Our proposed approach demonstrates strong potential in accurately detecting and localizing brain cancer.