Asan Medical Center secured multiple top finishes and hosted a major diagnostic event at the International Conference on Medical Image Computing and Computer-Assisted Intervention, known as MICCAI 2026, held over five days in Strasbourg, France, news.amc.seoul.kr reported.
This evaluation featured tasks in cavity segmentation and tuberculosis diagnosis on chest X-rays, drawing participation from 83 teams across 16 countries. Springer's Lecture Notes in Computer Science accepted all 11 final submitted papers from the event.
Asan Medical Center Researchers Win Top Awards
- What happened: Asan Medical Center researchers won top international awards for medical imaging AI and hosted the first global tuberculosis AI diagnostic challenge at MICCAI 2026.
Deep Learning Model Overcomes Low Resolution Data
Research teams from Asan Medical Center swept multiple technical tracks across separate imaging challenges at the conference. Their deep learning model utilized third-world pediatric brain MRI data, applying loss functions accounting for left-right brain asymmetry and dual-label learning strategies to overcome the low resolution and contrast inherent to portable, ultra-low-field MRI devices.
The team's algorithm maintained high performance even when input measurement data was reduced by 10 to 50 times, successfully generalizing to unseen environments and new anatomical regions.
Automated Chest CT Report Generation
The competition evaluated AI systems trained on approximately 47,000 chest CT scans and matching reports from nearly 20,000 patients contributed by 130 international teams.
The Asan research team combined a 3D medical image encoder with two auxiliary expert models to feed a large language model, curbing AI hallucination risks. Their system achieved first place in the Bilingual Evaluation Understudy metric, which measures sentence similarity between AI-generated findings and conclusions and actual physician-written reports.
| Competition Track | Asan Medical Center Ranking | Core Technical Focus |
|---|---|---|
| Tuberculosis Diagnostic AI Challenge (TREAT-MMTB) | Host Organizer | Chest X-ray cavity segmentation and tuberculosis classification |
| Low-field Pediatric Brain MRI Segmentation (LISA) | 1st Place | Ultra-low-field portable MRI automated structural analysis |
| 4D Flow MRI Image Reconstruction | 1st Place (All 3 Accuracy Categories) | Hemodynamic quantification and cross-institution generalization |
| 3D Vision-Language Modeling (VLM3D) | 2nd Place (1st in BLEU Sentence Similarity) | Automated chest CT report generation using vision-language models |
References
- International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI 2026) Conference Proceedings.
- Springer Lecture Notes in Computer Science (LNCS) – TREAT-MMTB Challenge Publications.
