Deep Predictive Learning of Carotid Stenosis Severity

TL;DR

This paper develops and compares machine learning models, including neural ODEs, to predict carotid stenosis severity, achieving about 77% accuracy, highlighting the potential and limitations of current data and methods.

Contribution

It introduces the application of Augmented Neural ODEs for stenosis severity prediction and systematically analyzes parameter effects, advancing deep learning approaches in medical diagnostics.

Findings

Neural ODEs improve prediction accuracy over classic methods.

Current SRUC data may be insufficient for high-precision predictions.

Achieved approximately 77% accuracy in severity classification.

Abstract

Carotid artery stenosis is the narrowing of carotid arteries, which supplies blood to the neck and head. In this work, we train a model to predict the severity of the stenosis blockage based on SRUC criteria variables and other patient information. We implement classic machine learning methods, decision trees and random forests, used in a previous experiment. In addition, we improve the accuracy through the use of the state-of-art Augmented Neural ODE deep learning method. Through systematic and theory-rooted analysis, we examine different parameters to achieve an accuracy of about 77%. These results show the strong potential in applying recently developing deep learning methods, while simultaneously suggesting that the current data provided by the SRUC criteria may be insufficient to predict stenosis severity at a high performance level.