Brain imaging data don’t improve Parkinson’s prediction models
Clinical assessments beat brain scans for forecasting long-term trajectory
Written by |
- Brain imaging data do not improve computer models used to predict long-term Parkinson’s disease progression.
- Incorporating MRI scans and machine learning models sometimes reduced the accuracy of motor symptom predictions.
- A small set of baseline clinical assessments remains the most effective way to forecast long-term cognitive and motor trajectories.
Brain imaging data are not very useful for computer models predicting the long-term trajectory of Parkinson’s disease, a study showed.
Researchers found that computer models aimed at predicting Parkinson’s progression were not meaningfully better when imaging data were included. In fact, some models became less accurate when imaging data were added.
The findings suggest that the best way to predict long-term Parkinson’s trajectories is with a small set of clinical assessments, the researchers said.
The study, “Baseline clinical features outperform structural MRI in predicting rapid cognitive and motor decline in Parkinson’s disease,” was published in npj Parkinson’s Disease.
Parkinson’s is a neurological disease that causes both motor and nonmotor symptoms. The disease is progressive, meaning these symptoms tend to worsen over time. But Parkinson’s doesn’t progress at the same rate for everyone, and there isn’t a reliable way to predict which patients will experience faster disease progression.
Training machines
Machine learning is a type of artificial intelligence that works by feeding data into a computer, which uses mathematical methods to identify patterns in the data. The patterns the computer learns can then be used to interpret or make predictions about other datasets.
Researchers have been exploring whether machine learning could be used to predict outcomes in diseases like Parkinson’s. Since machine learning is highly dependent on the data used to train the model, one key question is which type of data yields the best results.
“Prior studies have demonstrated that [machine learning] can predict cognitive outcomes in [Parkinson’s disease],” the researchers wrote. Their study, they said, “addresses a complementary question: whether [imaging data from MRI scans] provide incremental prognostic value.”
The team used several established datasets to train and test machine learning models to predict whether patients would have rapid progression of motor or nonmotor symptoms, as measured by standardized clinical assessments. They trained the models using two types of data: clinical assessments and brain imaging.
The scientists found that models trained solely on clinical assessments generally performed well, and that adding imaging data did not improve their ability to predict rapid cognitive or motor decline.
For nonmotor symptoms, models trained solely on clinical assessments were comparable to those trained on both clinical assessments and imaging data. For motor symptoms, models trained using both clinical assessments and imaging were less accurate than models that only used clinical assessments.
“These findings suggest that a small set of domain-specific clinical features is sufficient for early prognostic stratification, and that incorporating irrelevant features provides no additional benefit and may actively degrade performance,” the researchers wrote.
The team called for further work to continue refining how cutting-edge computer systems may help improve Parkinson’s care.
Leave a comment
Fill in the required fields to post. Your email address will not be published.