New hybrid CNN-Transformer model enhances retinal OCT classification with safety features

Researchers have developed a novel hybrid model combining a convolutional neural network (CNN) and a Transformer architecture for classifying retinal optical coherence tomography (OCT) scans. This model incorporates a gradient-boosting classifier and a three-part clinical safety layer designed to ensure the reliability of its confidence scores. The system achieves high accuracy while significantly reducing calibration error, making it the first OCT classifier to jointly validate confidence calibration, out-of-distribution rejection, and uncertainty flagging with publicly available weights. AI

IMPACT Enhances reliability of AI diagnostics in healthcare by improving confidence calibration and uncertainty flagging.

RANK_REASON The cluster describes a new academic paper detailing a novel model architecture and its evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• Calibrated Hybrid CNN-Transformer
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Retinal OCT Classification
• ScienceCast
• XGBoost

• paper
• model release
• safety

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