ChestX6-SSL-Benchmark Checkpoints
This repository hosts the pretrained and fine-tuned model checkpoints accompanying the manuscript:
"A Deployment Risk Score Framework for Self-Supervised Chest X-Ray Classification: Calibrated Multi-Objective Evaluation Under Annotation Scarcity and Scanner Heterogeneity"
Authors: Kashif Mahmood, Romana Aziz, Muhammad Ramzan, Mahwish Ilyas, and Ala Saleh Alluhaidan
Overview
The checkpoints correspond to the experiments performed on the ChestX6 dataset and include:
- Four supervised baseline models
- SimCLR self-supervised pretrained encoder
- MAE self-supervised pretrained encoder
- Fine-tuned SimCLR models
- Fine-tuned MAE models
Models are provided for 10%, 20%, and 100% labeled-data settings with three random seeds for each configuration.
Repository Structure
1supervised/
2 ResNet50 (Scratch)
3 ResNet50 (ImageNet)
4 EfficientNet-B0
5 MobileViT-XS
6
7ssl/
8 SimCLR pretrained encoder (.pth)
9 MAE pretrained encoder (.pth)
10
11finetuned/
12 SimCLR fine-tuned models (.pt)
13 MAE fine-tuned models (.pt)
Dataset
The experiments use the ChestX6: Multi-Class X-ray Dataset.
Original dataset:
The original dataset contains 18,036 images. During preprocessing, 48 duplicate images were removed, resulting in 17,988 unique images used in all experiments.
Cross-dataset evaluation uses ChestMNIST through the MedMNIST package.
Loading
The checkpoints store PyTorch model weights.
Refer to the accompanying GitHub repository for the complete training pipeline, model definitions, configuration files, and loading examples.
GitHub Repository
The complete source code, training pipeline, notebooks, configuration files, dataset split definitions, and reproduction instructions are available on GitHub:
The GitHub repository includes:
- End-to-end training and evaluation notebooks
- Fixed train/validation/test split definitions
- Experiment results and manuscript figures
- Instructions for reproducing all experiments reported in the accompanying manuscript
Citation
If you use these checkpoints, please cite the associated manuscript.
1@unpublished{mahmood2026drs,
2 title = {A Deployment Risk Score Framework for Self-Supervised Chest X-Ray Classification: Calibrated Multi-Objective Evaluation Under Annotation Scarcity and Scanner Heterogeneity},
3 author = {Kashif Mahmood and Romana Aziz and Muhammad Ramzan and Mahwish Ilyas and Ala Saleh Alluhaidan},
4 note = {Manuscript under revision},
5 year = {2026}
6}
License
MIT License.
