DefinePK

Abstract

Brain cancer is still a major worldwide health problem, and early and precise diagnosis may make a big difference in survival rates. Traditional diagnostic approaches that depend on manual MRI analysis take a lot of time, are subjective, and are easy to make mistakes, which mean they frequently miss modest tumor borders or early-stage malignancies. To overcome these constraints, this study presents an innovative hybrid deep learning system that integrates adaptive edge detection with dual-path CNN architecture. The approach starts with preprocessing and augmentation of T1/T2/FLAIR sequences. An adaptive Canny-Sobel filter with dynamic thresholding gets rid of noise from artifacts and healthy tissues while getting high-precision tumor outlines. A ResNet-50 backbone extracts hierarchical features from these edge maps and raw scans at the same time. A spatial attention module then enhances the outlines of the tumors. The suggested system has an average F1-score of 96.7% on a Kaggle dataset including 1,311 MRI scans during five-fold cross-validation. It has very high accuracy for glioma (100%) and recall for "no tumor" (98.67%). The suggested method gives radiologists a diagnostic tool that is easy to use and works in real time, which moves cancer treatment precision forward.