Researchers analyzed several computer models that use deep learning and radiomics to predict recovery outcomes for people who have suffered an ischemic stroke. These models look at specific markers in CT and MRI scans to estimate how well a patient might perform on the modified Rankin Scale after 90 days.
The study found that these computer models performed well in predicting outcomes. Interestingly, models that included clinical information along with imaging data performed significantly better than those that used only images. The study also found that models using CT scans performed similarly to those using MRI scans.
Because this was a meta-analysis of predictive models and not a clinical trial, these findings do not prove that the models will change standard care immediately. The results show that these tools have potential for predicting recovery, but more standardized ways to evaluate these models are still needed.
Common questions
How accurate are these computer models at predicting stroke recovery?
The study found that these models showed strong performance in predicting outcomes. Specifically, the internal validation showed a pooled AUC of 0.85, while external validation showed a pooled AUC of 0.74. These results indicate that the models have a solid ability to predict how a patient might perform 90 days after a stroke.
Do CT scans or MRI scans work better for these predictions?
The study found that models based on CT scans and those based on MRI scans showed similar performance. Both types of imaging were found to be comparable when used to predict the 90-day outcomes for patients who experienced an ischemic stroke.
Do these models work better when they include patient information?
Yes, the study found that models incorporating clinical variables significantly outperformed models that used only imaging data. Combining a patient's clinical information with their scan data provided a more effective way to predict their recovery outcomes.
