AI-Enhanced Multi-Model Medical Image Diagnosis: A Gated Multi-Scale Attention Framework for Brain and Lung Malignancy Detection

This research presents MS-GARAT v2, an innovative diagnostic framework built upon a DenseNet-18 backbone to detect malignancies in Brain MRI and Lung CT scans. By utilizing a Multi-Scale High-Frequency Information Retention (MS-HFIR) module and a Gated Adaptive Region Attention (G-ARA) mechanism, the model addresses the critical issue of textural biomarker erosion in deep networks. The system achieved a remarkable 99.65% accuracy for brain tumours and 99.48% for lung nodules, offering a computationally efficient and precise "second opinion" for automated oncology.