Researchers from the University of California-San Francisco are able to analyze MRI images using machine learning algorithms for computer-aided prognosis to predict short- and long-term outcomes for patients after acute spinal cord injuries.
In a proof-of-principle study, UCSF researchers used semi-automated image analysis with machine learning algorithms to assess the accuracy of axial T2-weighted radiomic features for classifying patients by degree of neurologic injury.
“We’ve been doing a lot of work looking at just imaging patterns on MRI in the spinal cord after injury,” says Jason Talbott, MD, assistant professor of radiology and biomedical imaging at UCSF, who adds that five machine learning algorithms were used to classify patients by degree of neurologic impairment with variable accuracy and identify potential prognostic texture features.
According to Talbott, researchers have been able to make sense of the various patterns and features detected on MRIs using the machine learning algorithms with a focus on the epicenter of acute spinal cord injuries, which are a major cause of morbidity and currently affect 282,000 Americans.
“The premise is that there’s a lot of information and data in the pattern and texture of the signal abnormality in the spinal cord after injury that we may not completely perceive or detect across patients over time just by looking at them,” Talbott observes. “We actually can use different machine learning algorithms to make sense out of all these different variables—more so than we can just by looking at the images on our own.”
Talbott will present the results of the study at the ARRS 2018 Annual Meeting, held in late April in Washington, DC. He contends that the machine learning algorithms act as a “prognostic biomarker” and that the accuracy rate in this “very preliminary study” was about 75 percent.
All five of the machine learning algorithms used in the study are freely available. However, going forward, Talbott says UCSF is working on some internally developed convolutional neural network algorithms that could increase the accuracy rate.
“We’re optimistic that by using this approach with refined algorithms, we can improve that accuracy,” he concludes. “If these algorithms can be effective in ultimately prospectively predicting patients’ outcomes and classifying them, that’s certainly a major advance. Spinal cord injury is a very heterogeneous pathology, and if you misclassify patients, you may miss effects on interventions and treatments.”
Talbott notes that these algorithms are being tested in a large prospective database that is being built as part of the Transforming Research and Clinical Knowledge in Spinal Cord Injury (TRACK-SCI) study.
Register or login for access to this item and much more
All Health Data Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access