Advancing Brain Tumor Detection: A Comparative Study of Densenet and Resnet Architectures in MRI Analysis

In evaluating brain malignancies, radiologists frequently look at several MRI sequences produced by multimodal imaging. Advanced MRI techniques that seek to correlate histological characteristics with radiological markers like cellular density or vascular structure have recently been the focus of neuro-oncology research. Currently, the most used imaging methods in clinical practice are T1-weighted sequences, which highlight anatomical characteristics, and T2-weighted sequences, which display oedema and can assist in determining cellularity. It is still challenging for radiologists to diagnose suspected gliomas or brain lesions on MRI using histological subtypes. However, because they usually rely on qualitative observations rather than predetermined quantitative thresholds, these diagnoses might be subjective. Without clear diagnostic thresholds, for instance, statements such as "low-to-moderate oedema could indicate tumor characteristics" are still ambiguous. As a result, tumor detection accuracy is still a major challenge, with existing techniques providing only mediocre reliability. AI has been a key component of neuro-oncology in recent years, with the potential to segment and categorize tumor subtypes in addition to detecting cancers in MR images. This study uses a DenseNet-based Convolutional Neural Network (CNN) model to improve brain tumor identification. The efficacy of pre-trained DenseNet and ResNet architectures in precisely detecting and describing brain cancers was assessed by comparing their performance under various settings.