NeuroXAI: Lightweight explainable early Alzheimer's detection for resource-constrained clinical settings
Medical Imaging 2026: Imaging Informatics, Vancouver, Kanada, 17 - 19 Şubat 2026, cilt.13930, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 13930
- Doi Numarası: 10.1117/12.3088200
- Basıldığı Şehir: Vancouver
- Basıldığı Ülke: Kanada
- Anahtar Kelimeler: Alzheimer's disease, clinical decision support, computeraided diagnosis, EfficientNetV2B0, explainable AI, Grad-CAM++, medical imaging informatics
- İstanbul Medipol Üniversitesi Adresli: Evet
Özet
Early detection of Alzheimer's disease (AD) remains limited in resource-constrained clinical settings due to the high cost of specialized analysis and lack of neuroimaging expertise. We present NeuroXAI, a clinical decision support system that integrates lightweight deep learning classification with real-time explainability for automated early AD detection from structural MRI. The system combines an EfficientNetV2B0 backbone achieving 88% accuracy on tri-class classification (cognitively normal, early mild cognitive impairment, and late mild cognitive impairment) with a multi-scale attribution framework generating Grad-CAM++, Guided Grad-CAM++, and consensus visualizations. NeuroXAI provides an integrated clinical interface supporting medical image viewing, interactive analysis, and explainability visualization. The lightweight architecture enables processing on standard clinical hardware without specialized GPU infrastructure. The system addresses the critical gap between research AI and clinical deployment by providing transparent, anatomically-grounded decision support for early cognitive impairment detection.