Rad-Linter achieves 98.5% accuracy in X-ray report error detection
November 20, 2026
Rad-Linter uses a 1.5B-parameter LLM fine-tuned with LoRA to check chest X-ray reports against image-derived pathology scores. It reaches 98.5% pair accuracy, compared to 80.5% for a keyword-and-negation rule baseline.
HOW THIS AFFECTS YOU
●
builderYou can use a small 1.5B model fine-tuned with LoRA for specialized medical error detection tasks.
●
healthThis tool can help doctors flag inconsistencies between X-ray images and written reports.