FM-Bridge: Feature Modulation Bridge for Liver Tumor Classification

Project Structure

• dataset.py - Custom Dataset class for loading and preprocessing 3D medical images
• model.py - Model definition with co-attention mechanisms
• utils.py - Utility functions for training and evaluation
• learnable.py - Text encoder and prompt learning classes
• train.py - Main training script

Dataset Structure

The expected data structure is:

./Dataset/
├── Internal_nii2/
│   └── <hospital_number>/
│       └── ap.nii.gz
├── Internal_roi/
│   └── <hospital_number>/
│       └── ap.nrrd
├── label_v2.csv
└── rad_feat.csv

Training

To train the model:

python train.py --text_prompt_len 6 --vision_prompt_dep 10 --vision_prompt_len 10 --orthogonal_loss_weight 0.1