OBJECTIVES
- Automated Precision: Develop a prompt-driven system using Segment Anything Model (SAM) for BraTS 2021 dataset.
- Domain Adaptation: Implementation of specialized bottleneck adapters for high-fidelity medical imaging features.
- Uncertainty Guidance: Integration of Monte Carlo Dropout to identify and refine ambiguous segmentation boundaries.
- Clinical Intelligence: Automated clinical report generation and visualization for immediate diagnostic support.
WORKFLOW
BraTS 2021 Data Pipeline
SAM Adapter Training
Auto-Prompt Generation
Uncertainty Refinement Loop
Clinical Insights Dashboard
RESULT
MedAdapt-SAM: Dashboard
Upload T1CE Patient MRI
Slice
(256x256 PNG/NIfTI)
(256x256 PNG/NIfTI)
Select Model: SAM-Adapter
(Auto)
PREDICTION:
GLIOMA
Confidence: 94.2%
Whole Tumor
0.91 DICE
Enhancing
0.88 DICE
AI Clinical Report:
High contrast enhancement noted in the right frontal lobe. Morphology suggestive of glioblastoma. Volume: 42.3cm³.
High contrast enhancement noted in the right frontal lobe. Morphology suggestive of glioblastoma. Volume: 42.3cm³.
3
Good Health
& Well-Being
& Well-Being
Social Benefits & Future Scope
MedAdapt-SAM contributes directly to SDG 3 by providing AI-assisted diagnostic tools that enhance the precision of brain tumor detection. This reduces diagnostic latency and empowers clinicians with data-driven reports, ultimately improving patient survival rates and well-being.