Researchers analyzed ultrasound images of carotid arteries in 666 patients to see if computer models could predict stroke risk. They compared advanced deep learning models against conventional machine learning methods like support vector machines and logistic regression. The goal was to find a better way to assess the danger of plaque buildup in the neck arteries.
The deep learning model, called ResNet-50, showed strong performance. It achieved an accuracy of 0.925 and an area under the curve of 0.982 for predicting stroke risk. This was significantly better than the conventional machine learning models tested in the study.
However, this was a retrospective study, meaning the data was looked at after the events occurred. The researchers noted that the accuracy of these assessments is still limited by how doctors subjectively interpret plaque features. Because of this, the results should be viewed as promising but not yet ready for immediate clinical use.
While deep learning models showed potential, doctors should not rely on them alone yet. The study highlights a need for better tools to objectively measure plaque vulnerability. Until these limitations are solved, standard clinical judgment remains essential for patient care.