Boosting Few-Shot Learning with Disentangled Self-Supervised Learning and Meta-Learning for Medical Image Classification [ In progress...]

Official code for Boosting Few-Shot Learning with Disentangled Self-Supervised Learning and Meta-Learning for Medical Image Classification. The code is based on DeepBDC by FeiLong, pytorch code of Joint Distribution Matters: Deep Brownian Distance Covariance for Few-Shot Classification and on IP-IRM by Wangt-CN, pytorch implementation of Self-Supervised Learning Disentangled Group Representation as Feature.

Dataset

We utilized the PI-CAI and BreakHis datasets for our experiments. To see pre-processing details, please refer to our paper. Based on our code, the data should be organized according to the following structure:

├── dataset
│   └── picai
│       ├── supervised                            
│       ├── unsupervised
│       ├── csv_files
│   └── breakhis
│       ├── supervised                            
│       ├── unsupervised
│       ├── csv_files

Here, supervised contains the samples used for supervised training, unsupervised the samples for the unsupervised pre-training steps, and csv_files the CSV files from which to retrieve the sample metadata.

Citation

@article{pachetti2024boosting,
  title={Boosting Few-Shot Learning with Disentangled Self-Supervised Learning and Meta-Learning for Medical Image Classification},
  author={Pachetti, Eva and Tsaftaris, Sotirios A and Colantonio, Sara},
  journal={arXiv preprint arXiv:2403.17530},
  year={2024}
}