Related Experiment Video
Updated: Sep 14, 2025
13:51
Cross-Modal Multivariate Pattern Analysis
Published on: November 9, 2011
20.1K
Large-vocabulary forensic pathological analyses via prototypical cross-modal contrastive learning
Chen Shen1, Chunfeng Lian2,3, Wanqing Zhang1
1Key Laboratory of National Ministry of Health for Forensic Sciences, School of Medicine & Forensics, Health Science Center, Xi'an Jiaotong University, Xi'an, Shaanxi, China.
Nature Communications
|July 23, 2025
Summary
A new visual-language model, SongCi, enhances forensic pathology by improving accuracy and efficiency in determining cause of death. This AI tool matches expert pathologist performance, addressing current field challenges.
Area of Science:
- Forensic Pathology
- Computational Pathology
- Artificial Intelligence
Background:
- Forensic pathology determines cause of death via post-mortem examinations.
- Current challenges include outcome variability, labor intensity, and professional shortages.
- Advanced computational tools are needed to support forensic analysis.
Purpose of the Study:
- Introduce SongCi, a visual-language model for forensic pathology.
- Enhance accuracy, efficiency, and generalizability of forensic analyses.
- Leverage prototypical cross-modal self-supervised contrastive learning.
Main Methods:
- Trained SongCi on a large multi-center dataset (16M+ image patches, 2,228 vision-language pairs).
- Validated performance against existing multi-modal and foundation models.
- Assessed model capabilities against forensic pathologists with varying experience levels.
Main Results:
- SongCi demonstrated superior performance in forensic tasks compared to existing models.
- The model matched the capabilities of experienced forensic pathologists.
- SongCi significantly outperformed less experienced practitioners.
Conclusions:
- SongCi offers a powerful AI solution to support forensic pathology.
- The model improves diagnostic accuracy and efficiency.
- SongCi provides robust multi-modal explainability for forensic findings.
