The advent of foundation models (FMs) in healthcare offers unprecedented opportunities to enhance medical diagnostics through automated classification and segmentation tasks. However, these models also raise significant concerns about their fairness, especially when applied to diverse and underrepresented populations in healthcare applications. Currently, there is a lack of comprehensive benchmarks, standardized pipelines, and easily adaptable libraries to evaluate and understand the fairness performance of FMs in medical imaging, leading to considerable challenges in formulating and implementing solutions that ensure equitable outcomes across diverse patient populations. To fill this gap, we introduce FairMedFM, a fairness benchmark for FM research in medical imaging. FairMedFM integrates with 17 popular medical imaging datasets, encompassing different modalities, dimensionalities, and sensitive attributes. It explores 20 widely used FMs, with various usages such as zero-shot learning, linear probing, parameter-efficient fine-tuning, and prompting in various downstream tasks -- classification and segmentation. Our exhaustive analysis evaluates the fairness performance over different evaluation metrics from multiple perspectives, revealing the existence of bias, varied utility-fairness trade-offs on different FMs, consistent disparities on the same datasets regardless FMs, and limited effectiveness of existing unfairness mitigation methods.
FairMedFM captures comprehensive modules for benchmarking the fairness of foundation models in medical image analysis.
- Dataloader: provides a consistent interface for loading and processing imaging data across various modalities and dimensions, supporting both classification and segmentation tasks.
- Model: a one-stop library that includes implementations of the most popular pre-trained foundation models for medical image analysis.
- Usage Wrapper: encapsulates foundation models for various use cases and tasks, including linear probe, zero-shot inference, PEFT, promptable segmentation, etc.
- Trainer: offers a unified workflow for fine-tuning and testing wrapped models, and includes state-of-the-art unfairness mitigation algorithms.
- Evaluation includes a set of metrics and tools to visualize and analyze fairness across different tasks.
| Tasks | Supported Usages | Supported Models | Supported Datasets |
|---|---|---|---|
| Image Classification | Linear probe, zero-shot, CLIP adaptaion, PEFT | CLIP, BLIP, BLIP2, MedCLIP, BiomedCLIP, PubMedCLIP, DINOv2, C2L, LVM-Med, MedMAE, MoCo-CXR | CheXpert, MIMIC-CXR, HAM10000, FairVLMed10k, GF3300, PAPILA, BRSET, COVID-CT-MD, ADNI-1.5T |
| Image Segmentation | Interactive segmentation prompted with boxes and points | SAM, MobileSAM, TinySAM, MedSAM, SAM-Med2D, FT-SAM, SAM-Med3D, FastSAM3D, SegVol | HAM10000, TUSC, FairSeg, Montgomery County X-ray, KiTS, CANDI, IRCADb, SPIDER |
-
Release the classification tasks.
-
Release the segmentation tasks.
- 2D dataset + 2D SAMs
- 3D dataset + 2D SAMs
- 3D dataset + 3D SAMs
-
Release more models
-
Release the preprocessed datasets.
-
Integration of the classic strategies.
-
Release examples and tutorials.
The installation requires three steps.
-
Download from github
git clone https://github.com/FairMedFM/FairMedFM.git cd FairMedFM -
Creating conda environment
conda env create -f environment.yaml conda activate fairmedfm -
Download Pretrained FMs
wget https://object-arbutus.cloud.computecanada.ca:443/rjin/pretrained.zip unzip pretrained.zip rm -f pretrained.zip
Our notebook tutorials also contains how to setup the environment in Colab.
You can either download our pre-processed data directly (see next section) or pre-process customized data your self. However, not all dataset we used permit us to release the data on our end (e.g., dataset like MIMIC and ADNI requires the user go through their data usage application first). In such case, we cannot provide the download link of our preprocessed dataset for them, but we have the original dataset downloading link and our pre-process scripts released.
We provide data preprocessing scripts for each datasets here. The data preprocessing contains 3 steps:
- (Optional) preprocess imaging data.
- Preprocess metadata and sensitive attributes.
- Split dataset into training set and test set with balanced subgroups (for classification only).
Our data is downloaded uisng the following links.
| Dataset | Link |
|---|---|
| CheXpert | Original data Demographic data |
| MIMIC-CXR | MIMIC-CXR |
| PAPILA | PAPILA |
| HAM10000 | HAM10000 |
| OCT | OCT |
| OL3I | OL3I |
| COVID-CT-MD | COVID-CT-MD |
| ADNI | ADNI-1.5T |
| Dataset | Link |
|---|---|
| HAM10000 | HAM10000 |
| TUSC | TUSC |
| FairSeg | FairSeg |
| Montgomery County X-ray | Montgomery County X-ray |
| KiTS2023 | KiTS2023 |
| IRCADb | IRCADb |
| CANDI | CANDI |
| SPIDER | SPIDER |
We offer data downloading through the S3 link. We are working to build this feature now.
| Dataset | Link |
|---|---|
| CheXpert | Requires application on original data provider. |
| MIMIC-CXR | Requires application on original data provider. |
| PAPILA | PAPILA |
| HAM10000 | HAM10000 |
| OCT | TODO |
| OL3I | TODO |
| COVID-CT-MD | TODO |
| ADNI | Requires application on original data provider. |
We offer some examples of how to use our package through the notebook.
| Feature | Notebook |
|---|---|
| Linear Probing |  |
| CLIP Zero-shot and Adaptor | |
| Segmentation | |
We provide an example of running a linear-probe (classification) experiment of the CLIP model on the MIMIC-CXR dataset to evaluate fairness on sex. Please refer to parse_args.py for more details.
python main.py --task cls --usage lp --dataset CXP --sensitive_name Sex --method erm --total_epochs 100 --warmup_epochs 5 --blr 2.5e-4 --batch_size 128 --optimizer adamw --min_lr 1e-5 --weight_decay 0.05We also provide an example of using SAM with center point prompt on the TUSC dataset to evaluate fairness on sex. Please refer to parse_args.py for more details.
python main.py --task seg --usage seg2d --dataset TUSC --sensitive_name Sex --method erm --batch_size 1 --pos_class 255 --model SAM --sam_ckpt_path ./weights/SAM.pth --img_size 1024 --prompt centerWe thank MEDFAIR for their pioneering works on benchmarking fairness for medical image analysis, and Slide-SAM for the SAM inference framework.
Guidelines for Responsible Use of the Benchmark
- Comprehend the Benchmark's Design: Before utilizing the benchmark, ensure a thorough understanding of its design, including the datasets, models, tasks, and metrics involved. Be aware of its intended scope, as well as any limitations or biases inherent in the datasets and models.
- Acknowledge: The benchmark uses publically available data, where some of the data needs users' license in order to download. We acknowledge that the these data belongs to the original owner, where FairMedFM uses them under the constraint set by these owners.
- Fairness and Bias Awareness: The benchmark includes fairness metrics to assess the performance of AI models. Users should carefully consider these metrics to avoid perpetuating or amplifying biases in AI systems, especially in sensitive domains like healthcare.
- Use the Provided Codebase: The benchmark comes with a codebase designed to facilitate reproducibility. Users are encouraged to utilize and contribute to this codebase to maintain a consistent standard of experimentation.
- Document Any Changes: If modifications are made to the benchmark, such as adjusting datasets or metrics, these changes should be well-documented and justified. This ensures that results can be accurately interpreted and compared.
- Engage with the Community: Users are encouraged to engage with the research community by sharing findings, discussing potential improvements, and collaborating on extensions of the benchmark.
- Attribution and Acknowledgment: When using the benchmark in research or applications, proper attribution should be given to the creators of the benchmark and the original datasets. Acknowledge the sources of any third-party data or models used.
- Feedback and Contributions: Provide feedback to the benchmark’s maintainers regarding any issues or potential improvements. Contributions in the form of new datasets, metrics, or models are encouraged to enhance the benchmark's utility and relevance.
- Stay Updated: Keep abreast of updates to the benchmark or related research. This ensures that you are using the most current and validated version, which may include important improvements or corrections.
By following these guidelines, users can ensure that they are utilizing the benchmark responsibly, contributing to ethical AI development, and fostering a collaborative research environment.
This project is released under the CC BY 4.0 license. Please see the LICENSE file for more information.
@article{jin2024fairmedfm,
title={FairMedFM: Fairness Benchmarking for Medical Imaging Foundation Models},
author={Jin, Ruinan and Xu, Zikang and Zhong, Yuan and Yao, Qiongsong and Dou, Qi and Zhou, S Kevin and Li, Xiaoxiao},
journal={arXiv preprint arXiv:2407.00983},
year={2024}
}