Multimodal Feature Prototype Learning for Interpretable and Discriminative Cancer Survival Prediction

paper

💡Primary contributions

To overcome the challenges of interpretability in survival analysis, we propose FeatProto. This is a novel framework for Feature Prototype Learning in WSI-Genetic based survival prediction. Our key contributions are summarized as follows:

1. 🕐 The first design of the Feature Prototype Library with Wandering Prototypes.
2. 🕐 EMA ProtoUp: Exponential Moving Average Prototype update strategy.
3. 🕐 MPMatch: Multi-level Deep Prototype Matching Strategy.
4. 🕐 ProtoSurv Loss: Prototype-Survival Fusion Loss.

🧗 Compare with other survival prediction methods

Feature prototype learning with global-local fused features and multi-level deep prototype matching for interpretable survival prediction;

💡highlight

📃 Dataset

Download two public datasets from the following link:

TCGA-LUAD

TCGA-BLCA

📃 Preprocess

The CLAM model was employed for fully automated data preprocessing.

📝Requirements

Prepare the available PyTorch(>=2.5.1) environment and ：

pip install -r requirements.txt

🔥 Training

To train and test our model, please run the following command:

python main.py

📋 Before training, specify the dataset and training configuration using the "options_XX_XX.py" file in the "utils" folder.
📋 Other experimental files can be found in the "experiment" folder.