Akademik Veri Yönetim Sistemi
23rd IEEE International Symposium on Biomedical Imaging, ISBI 2026, London, İngiltere, 8 - 11 Nisan 2026, cilt.2026-April, (Tam Metin Bildiri)
Deep learning models achieve significant diagnostic effectiveness in medical imaging but remain hindered by their lack of explainability. This study presents ClinAlignNet, a clinician-interactive and fidelity-driven explainability framework for brain MRI analysis. The method combines a residual counterfactual mechanism with alignment-based regularisation and uncertainty estimation to produce explanations that are anatomically coherent and clinically faithful. Evaluated on fifty patient cases from the Children's Brain Tumour Network, ClinAlignNet demonstrates superior performance to Grad-CAM and LIME with higher Dice (0.8874), Jaccard (0.8973), XAlign (0.9285), and Explanation Fidelity Score (0.896), while achieving a lower Hausdorff distance (23.96). Qualitative assessment by clinicians and a brain surgeon confirmed that the generated explanations align strongly with tumour boundaries and expert annotations. The framework bridges the gap between algorithmic transparency and clinical reasoning, offering a reliable pathway for the adoption of explainable AI in neuro-oncological diagnosis.