3D Vision Transformers & Deep CNNs for Acute Stroke Lesion Segmentation and Early Oncological Detection
Published in Ongoing Research Project, 2026
Medical Imaging & Healthcare AI 3D ViT & CNN Architectures Grad-CAM & SHAP Interpretability Ongoing Research
Research Objectives
Medical imaging AI demands extraordinary precision paired with clinician-trusted interpretability to be safely deployed in high-stakes triage and diagnostics.
- Acute Ischemic Stroke Lesion Segmentation: Designing volumetric 3D Vision Transformers (ViT) with hybrid convolutional backbones for precise penumbra and ischemic core delineation on brain CT/MRI scans.
- Early Cancer Screening: Analyzing multi-view mammography and digital histopathology gigapixel whole slide images (WSI) for early microcalcification and neoplastic biomarker identification.
- Trustworthy Clinical Attribution: Formulating multi-scale Grad-CAM and voxel-level SHAP heatmaps to corroborate algorithmic predictions against radiologist and pathologist annotated regions of interest.
Recommended citation: S. M. Mohiuddin Khan Shiam. (2026). "3D Vision Transformers & Deep CNNs for Acute Stroke Lesion Segmentation and Early Oncological Detection." Working Paper / Ongoing Research.