Machine Learning May Help Predict Incomplete Stent Expansion

Deep learning may be able to predict incomplete stent expansion, new research in JACC: Cardiovascular Interventions suggests.

“Although post-stenting intravascular ultrasound (IVUS) has been used to optimize percutaneous coronary intervention (PCI), there are no pre-procedural guidelines to estimate the degree of stent expansion and provide preemptive management before stent deployment,” the authors noted in their abstract.

The research team, looking to develop pre-procedural IVUS-based models to predict stent underexpansion, examined a total of 618 coronary lesions in 618 patients undergoing PCI were randomly assigned to training and test sets (5:1 ratio). Pre- and post-stenting IVUS images were obtained (along with clinical information such as stent diameter, length, and inflation pressure; balloon diameter; and maximal balloon pressure), and the pre-procedural models used to develop a regression model using a convolution neural network to predict area post-stenting.

The researchers reported that the overall frequency of stent under-expansion was 15% (5,209 of 34,736 frames). Stent areas predicted by the pre-procedural ICUS-based regression model significantly correlated with stent areas measured with IVUS post-stenting (r=0.802). Maximal accuracy for the prediction of stent underexpansion was 94% when the convolutional neural network– and mask image–derived features were used for the classification model. There were significant correlations between predicted and measured minimal stent area (r=0.832) as well as predicted and measured total stent volume (r=0.958).

“Deep-learning algorithms accurately predicted incomplete stent expansion,” the researchers wrote. “A data-driven approach may assist clinicians in making treatment decisions to avoid stent underexpansion as a preventable cause of stent failure.”

Read the abstract here.