Domain: Healthcare/Radiology
Description: MIT studies reveal why AI models that analyze medical images can be biased, affecting diagnostic accuracy across demographic groups.

Ethical Challenges:

• Demographic Bias: Diagnostic algorithms show varying performance across racial and gender groups
• Training Data Imbalances: Underrepresentation of certain demographic groups in training datasets
• Clinical Decision Impact: Biased diagnostic tools can lead to misdiagnosis or delayed treatment
• Healthcare Equity: Perpetuation of existing healthcare disparities

Public Datasets:

• NIH Chest X-ray Dataset: 112,120 frontal-view X-ray images from 30,805 unique patients
• URL: https://www.nih.gov/news-events/news-releases/nih-clinical-center-provides-one-largest-publicly-available-chest-x-ray-datasets-scientific-community
• CheXpert Dataset: Stanford's large dataset of chest radiographs with labels
• URL: https://stanfordmlgroup.github.io/competitions/chexpert/
• RSNA Pneumonia Detection Challenge: Kaggle dataset with demographic annotations
• URL: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge