[MICCAI2024] Mask-Free Neuron Concept Annotation for Interpreting Neural Networks in Medical Domain

This repo is the official source code for 'Mask-Free Neuron Concept Annotation for Interpreting Neural Networks in Medical Domain' International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)

Introduction

Preparation

1. Create virtual environment by conda.

conda create -n MAMMI python=3.10
conda activate MAMMI
pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
pip install -r requirements.txt

• Note
  • MedCLIP: pip install git+https://github.com/RyanWangZf/MedCLIP.git
  • CLIP: pip install git+https://github.com/openai/CLIP.git

2. Prepare resources to run code.

   Data

   • Probing set: NIH14, ChestX-det (for visualization)
   • Concept set MIMIC-CXR Report; We provide preprocessed MIMIC CXR Report test data in this repo. ('./dataset/report')

   Pre-trained model
   Model(Link): DenseNet121(Moco v2), ResNet50(Moco v2)

1. Prepare Concept set (MIMIC Nouns)

run 'prepare_mimic_nouns.py'

• # of MIMIC Nouns = 1361

2. Example Selection

run 'example_selection.py'

3. Concept matching

run 'concept_matching.py'

Visualization

TBD