🎓 Data Scientist passionate about advancing AI-driven Medical Imaging.
🔬 I build and evaluate deep learning pipelines for explainable diagnosis and quantitative MRI biomarker estimation.
|
Programming & Data Science
|
Neuroimaging & MRI Tools |
- Built an end-to-end PyTorch pipeline for pediatric pneumonia detection from chest X-rays using a fine-tuned ResNet-18 backbone.
- Achieved AUROC = 0.979, AUPRC = 0.985, with Sensitivity = 0.997 (screening) and Precision = 0.94 (rule-in mode).
- Applied Grad-CAM visualizations to highlight pulmonary opacities, ensuring transparency and clinical interpretability.
- Implemented temperature scaling and Youden’s J threshold tuning for probability calibration and operating-point flexibility.
- Emphasized explainability, robust evaluation, and reproducible design for research-grade medical imaging AI.
- Built a 3D U-Net conditional GAN to synthesize CMRO₂ maps from multimodal quantitative MRI (CBV, CBF, T2, T2*).
- Preprocessing included GM masking, MNI152 resampling, and normalization.
- Achieved high accuracy with all modalities (MSE ~0.0007, SSIM ~0.92, Pearson r ~0.95).
- Demonstrated that vascular modalities (CBV & CBF) are essential for reliable CMRO₂ estimation.
- Implementing selected Fabber forward models in pure Python, focusing on Bayesian estimation for multi-echo (TE) and multi-inversion time (TI) ASL data.
- Validating against synthetic and age-related cohorts using voxel-/ROI-wise comparisons.
- Extending modeling to evaluate fitted vs. fixed T2 values, improving robustness across populations.
✨ “Advancing healthcare through imaging, AI, and biomedical engineering.”