Related Experiment Video
Updated: Jul 3, 2026
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
An expert-level vision-language model for multitask diagnostic morphology in clinical laboratories
NPJ Digital Medicine
|June 22, 2026
Summary
Lingjian, a novel vision-language model, enhances clinical laboratory morphology analysis by accurately identifying and interpreting cell images. This AI tool significantly improves diagnostic accuracy and efficiency, outperforming human experts in assessments.
Area of Science:
- Artificial Intelligence in Medicine
- Computational Pathology
- Medical Imaging Analysis
Background:
- Microscopic morphology analysis in clinical labs is complex, subjective, and hard to scale.
- Existing methods lack efficiency and consistency, posing challenges for accurate diagnosis.
Purpose of the Study:
- To introduce Lingjian, a vision-language model specifically designed for laboratory morphology.
- To enhance cell identification, morphology description, interpretation, and localization using AI.
Main Methods:
- Developed Lingjian on Qwen3-VL-8B, utilizing multistage domain adaptation.
- Trained on over 400,000 laboratory images with text and grounding supervision.
- Evaluated on public benchmarks, cross-domain test sets, and clinical external quality assessments.
Main Results:
- Lingjian achieved 93.0% accuracy on the National Center for Clinical Laboratories EQA (2021-2025), surpassing human experts (78.1%) and general models (75.3%).
- In reader studies, Lingjian assistance boosted junior readers' sensitivity for abnormal screening from 69.7% to 91.7% with high specificity.
- Demonstrated robust multitask performance in image comprehension, cell identification, and interpretation.
Conclusions:
- Lingjian offers a scalable and accurate solution for clinical laboratory morphology.
- The model shows potential to significantly improve diagnostic workflows and reduce diagnostic errors.
- Released model weights and resources encourage further research and application in medical AI.
