New TRIAGE framework boosts OCT classification accuracy with less labeled data

Researchers have developed TRIAGE, a novel semi-supervised framework designed to improve the accuracy of retinal optical coherence tomography (OCT) classification. This method addresses the challenge of insufficient labeled data by employing a risk-controlled approach with an asymmetric cost matrix, which accounts for different types of classification errors. TRIAGE integrates a hierarchical classifier, a patient-grouped conformal risk controller, and a Transformer teacher for cross-slice verification. The framework has demonstrated significant performance gains, achieving high accuracy and AUC scores with minimal labeled data, and outperforming existing semi-supervised methods, particularly in reducing under-grading rates. AI

IMPACT This research could lead to more efficient and accurate AI-driven diagnostic tools for retinal diseases, reducing the need for extensive expert annotations.

RANK_REASON The cluster describes a new research paper detailing a novel framework for a specific classification task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Md Ashraful Hossen Akash
• Noor Eye Hospital
• OCT-C8
• optical coherence tomography
• Transformer++
• TRIAGE

• paper
• model release
• other

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