El-amin/FairCXRnet-A-Multi-Task-Learning-Model-for-Chest-X-Ray-Classification-for-Low-Resource-Settings

This is notebook isfrom a research paper "FairCXRnet: A Multi-Task Learning Model for Domain Adaptation in Chest X-Ray Classification for Low Resource Settings " # FairCXRnet **FairCXRnet: A Multi-Task Learning Model for Domain Adaptation in Chest X-Ray Classification for Low Resource Settings** --- ## Overview This repository contains the official implementation of airCXRnet, a multi-task deep learning model designed for domain adaptation in chest X-ray (CXR) classification. The model addresses challenges in low-resource settings by improving generalization, fairness, and robustness** across underrepresented populations and diverse imaging domains. FairCXRnet integrates: - **Multi-task learning** (classification + domain alignment tasks) - **Domain adaptation** strategies to handle dataset shift - **Fairness-aware components** for equity across demographic groups This code accompanies our paper: > **FairCXRnet: A Multi-Task Learning Model for Domain Adaptation in Chest X-Ray Classification for Low Resource Settings** > *Aminu Musa, Rajesh Prasad, Mohammed Hassan, Mohamed Hamada, Saratu Yusuf Ilu* > (2025, ETLTC Conference, Japan) --- ## ⚙️ Features - Multi-task training framework with **shared backbone + task-specific heads** - Support for **multi-label classification** (e.g., ChestMNIST, NIH CXR, local datasets) - Domain adaptation using **adversarial training & feature alignment** - Evaluation pipeline for **cross-domain generalization** - Fairness-aware analysis (demographic and dataset-level performance evaluation) --- ## 📂 Repository Structure FairCXRnet/ │── data/ # Dataset loading & preprocessing scripts │── models/ # Model architectures (FairCXRnet, DenseNet, etc.) │── training/ # Training loops, losses, domain adaptation strategies │── evaluation/ # Evaluation, fairness metrics, visualization │── utils/ # Helper functions │── configs/ # Config files for experiments │── main.py # Entry point to train/evaluate the model │── requirements.txt # Dependencies │── README.md # Project documentation (this file) FairCXRnet shows: --- 📊 Results Improved cross-domain generalization compared to single-task baseline …