metadata

language: en
license: mit
tags:
  - deep-learning
  - oncology
  - pathology
  - image-classification
model-index:
  - name: APEDIA - Angiosarcoma PD-L1 Expression Diagnostics
    results:
      - task:
          name: PD-L1 Tumor Proportion Score Prediction
          type: tps-prediction
        metrics:
          - name: Pathologist Correlation
            type: correlation
            value: 0.87

APEDIA Model Weights

Overview

This repository hosts the model weights for APEDIA (Angiosarcoma PD-L1 Expression DIAgnostics with Deep Learning), a deep learning tool designed to assist pathologists in assessing PD-L1 expression in angiosarcoma through the calculation of the Tumor Proportion Score (TPS) from whole-slide images (WSIs).

APEDIA aims to enhance decision-making quality in the pathology of angiosarcoma by automating the detection of tumor areas and the classification of cell types based on PD-L1 expression.

Model Checkpoints

This repository contains two critical checkpoints for the APEDIA pipeline:

tp_pred_model_checkpoint.pth: This model checkpoint is responsible for detecting tumor areas in PD-L1 stained Whole-Slide Image (WSI) patches.
seg_model_checkpoint.pth: This checkpoint powers the detection and classification of cell types within tumor patches, distinguishing between background, non-tumor cells, PD-L1 positive cells, and PD-L1 negative cells.
deepliif_latest_net_G1.pth: Derived from the DeepLIIF library, this is the checkpoint of a generative adversarial network (GAN) designed to isolate the hematoxylin component in IHC-stained slides. For more information on DeepLIIF, visit their GitHub repository.

Usage

The model weights will be automatically downloaded by the APEDIA library when executing predictions to calculate the Tumor Proportion Score (TPS). Ensure that you have the latest version of APEDIA installed to use these models effectively.

For detailed instructions on integrating these model weights with APEDIA and running the prediction pipeline, please refer to the APEDIA repository.

License

The model weights are available under the MIT License. For more details, see the LICENSE file in this repository.