Accuracy of IVUS-Based Machine Learning Segmentation Assessment of Coronary Artery Dimensions and Balloon Sizing
Authors' conclusion
ML model IVUS segmentation measurements were well-correlated with those of experts and selected an appropriate balloon size in more than 90% of images.
Study context
- Population
- Patients with intravascular ultrasound (IVUS) images of coronary arteries (IVUS-guided PCI optimization)test set: 437 images from 92 patients
- Comparison
- Machine learning (ML) automatic IVUS segmentation vs expert analysis
- Outcome
- Agreement rate between ML vs experts' measurements for appropriate balloon size selection, and lumen and acute stent areas (lumen area difference ≥0.5 mm² clinically significant)
Methodological detailsModel developed on 8,076 IVUS images from 234 patients, randomly split into training (83%) and validation (17%); evaluated on independent test set
- Model developed on 8,076 IVUS images from 234 patients, randomly split into training (83%) and validation (17%); evaluated on independent test set
Performance
As reported in the abstract
Agreement rate ML vs experts (test data set)
| Test or model | Agreement, lumen area | Agreement, acute stent area | Agreement, balloon size by vessel diameter only | Agreement, balloon size adding lumen diameter |
|---|---|---|---|---|
| ML segmentation | 85.5% | 97.0% | 70.6% | 92.4% |
ML segmentation
Agreement, lumen area85.5%
Agreement, acute stent area97.0%
Agreement, balloon size by vessel diameter only70.6%
Agreement, balloon size adding lumen diameter92.4%
Further findings
| Outcome | ML segmentation |
|---|---|
| Correlation with experts' segmentation (training data set) | |
| Correlation coefficient, lumen area | 0.992 |
| Correlation coefficient, vessel area | 0.993 |
- Correlation with experts' segmentation (training data set): training set: 8,076 images from 234 patients
Original publication
Accuracy of IVUS-Based Machine Learning Segmentation Assessment of Coronary Artery Dimensions and Balloon Sizing. JACC Adv. Published 22 August 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →