Advancing Medical Image Analysis with Feature Fusion Deep Networks for Disease Classification

Medical image classification is essential for diagnosing and treating serious diseases like lung and brain cancers since early detection of these complex formations improves patient outcomes. Current methods misclassify because they cannot catch subtle patterns and large-scale features. We solved this challenge with the Multiscale Transformer-enhanced Feature Fusion Network (MTFFN), an upgraded architecture to improve classification accuracy. We use a domain-specific transformer encoder and convolutional neural networks (MC-CNN) to extract fine-grained visual information and high-level semantics. By augmenting data to account for picture variability, MTFFN enhances generalization and diagnostic accuracy. Experimental validation on the BRATS and LIDC datasets shows that MTFFN outperforms state-of-the-art methods with 99.3% accuracy, 99.5% specificity, and 99.0% sensitivity. These findings show that MTFFN is reliable, especially for classifying complex tumor forms. MTFFN's impact on early detection and tailored treatment may grow with automated clinical procedures.