Using machine learning for detection of Parkinson's disease and mild cognitive impairment.

BACKGROUND

Parkinson's disease is a movement disorder featuring motor symptoms and cognitive decline, which can manifest as mild cognitive impairment. The incidence of mild cognitive impairment increases with disease progression, and Parkinson's disease can cause significant disability, therefore, identification of Parkinson's disease and mild cognitive impairment in Parkinson's disease is imperative. Neuroimaging and biofluid biomarkers have been studied separately, however, research suggests that combining biomarkers may improve detection.

OBJECTIVES

We aimed to investigate using machine learning whether a combination of neuroimaging and biofluid biomarkers would result in more effective identification of Parkinson's disease and mild cognitive impairment.

METHODS

Utilizing the Parkinson's Progression Markers Initiative dataset, we applied two different machine learning approaches, support vector machine and random forest, to explore combinations of neuroimaging and cerebrospinal fluid biomarkers for detection.

RESULTS

Overall, both machine learning techniques had an equivalent performance. In general, in those models for detecting Parkinson's disease, DaT-SPECT performed better than biofluid biomarkers. In models for detecting Parkinson's disease patients with mild cognitive impairment, combining DaT-SPECT with phosphorylated-tau-181 resulted in higher accuracy, outperforming DaT-SPECT alone.

CONCLUSIONS

Classification for Parkinson's disease and mild cognitive impairment may be improved by combining neuroimaging with biofluid biomarkers through machine learning models.