Coronary disease & ACSDiagnostic or prognostic study
Derivation and external validation of a deep learning model to predict changes in coronary plaque burden
- Derivation and external validation of a deep learning model
Authors' conclusion
This is the first deep learning model capable of detecting changes in plaque progression by analysing the rate of plaque burden change between adjacent frames.
Study context
- Population
- Coronary regions of interest assessed by intravascular ultrasound (IVUS) in the IBIS-4 and PACMAN-AMI datasetsIBIS-4 regions of interest (ROIs) (derivation and validation) 1,960 · PACMAN-AMI ROIs (testing, external validation) 5,283
- Participants
- 1,960 vs 5,283
- Comparison
- IBIS-4 vs PACMAN-AMI
- IBIS-4n = 1,960
- PACMAN-AMIn = 5,283
Further findings
Model performance, derivation and validation (IBIS-4)
- Accuracy, mean±standard deviation
- 0.85±0.02
- Matthews correlation coefficient
- 0.70±0.04
- F1 score for both progression and regression classes
- 0.85±0.02
ΔPAV in the PACMAN-AMI testing dataset
- Mean ΔPAV, all ROIs
- −0.31±5.63
- ROIs with regression
- 2,665
- Mean ΔPAV, regression ROIs
- −4.57±3.73
- ROIs with progression
- 2,618
- Mean ΔPAV, progression ROIs
- 4.02±3.55
- Plaque progression prevalence
- 49.6%
Model performance, testing (external validation, PACMAN-AMI)
- Accuracy
- 0.84
- Matthews correlation coefficient
- 0.68
- F1 score for progression and regression classes
- 0.84
- across the 100 trained models
Original publication
Derivation and external validation of a deep learning model to predict changes in coronary plaque burden. EuroIntervention. Published 4 May 2026. doi:10.4244/eij-d-25-01352
Summarised from the published abstract; every number is checked against it