Missing-Data-Robust Medical Imaging Model
Apr 2026
Finished
Medical model for robust survival prediction in glioblastoma. Stage 1: self-supervised pretraining (SSL) on the BraTS brain tumor segmentation dataset, learning generic 3D image representations. Stage 2: fine-tuning the pretrained encoder for survival time prediction using the UPenn-GBM dataset with multimodal clinical and imaging data (NIfTI). Stage 3: evaluation on predefined data partitions. Architecture uses PyTorch with automatic GPU acceleration, YAML-based configuration via dataclass, early stopping, automatic checkpointing, and deterministic seeding. Designed to handle missing data in real clinical environments.
Technologies
AI
Jupyter
LaTeX
NumPy
Pandas
Plotly
Python
PyTorch
Sklearn
