New SAGE dataset targets AI bias in South Asian medical imaging

Researchers have introduced SAGE, a new dataset of 1,300 expert-annotated GI endoscopy images from the South Asian region. This dataset aims to address the underrepresentation of diverse geographic populations in existing AI training data, which can lead to biased models. SAGE includes images, captions, hallucination tags, labels, and question-answer pairs, making it suitable for tasks like image captioning, classification, and fine-tuning large multimodal models (LMMs). Benchmarking with SAGE revealed significant performance drops in existing models when applied to South Asian data, highlighting the need for more inclusive AI development in healthcare. AI

IMPACT This dataset could help mitigate population bias in medical AI, leading to more equitable diagnostic tools for underrepresented regions.

RANK_REASON The cluster describes a new academic dataset release for multimodal learning and bias analysis in medical AI. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Europe
• LMMs
• Niyoj Oli
• SAGE

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →