This repository contains outputs from the clinical-only baseline survival model.
Only structured clinical variables were used for prediction—no MRI data.
This ablation experiment serves as a baseline for quantifying the incremental value contributed by MRI-derived features.
Contents
clinical_only_outputs.zip
models/ — best checkpoints for all 5 folds
predictions_fold*.csv — per-fold risk predictions and survival targets
Training Settings
5-fold cross-validation
Epochs: 30
Batch size: 2
Learning rate: 3e-5
Weight decay: 1e-5
Loss: Cox proportional hazards
Optimizer: Adam
Performance
Mean C-index:0.7967 ± 0.1067
Notes
This model provides a reference point for understanding how much predictive value is added when incorporating MRI imaging alongside clinical data.
For full training code and experiment setup, see: https://github.com/AliEngineering/multimodal-prostate-survival
