SPECT Image Features for Early Detection of Parkinson's Disease using Machine Learning Methods

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Parkinson's disease (PD) is a widespread neurodegenerative disorder lacking a cure.
• Early diagnosis of PD is crucial for effective intervention and patient management.
• Supervised machine learning (ML) models show promise for early PD diagnosis using clinical data, but performance varies with data and features.

Purpose of the Study:

• To evaluate the efficacy of 23 single photon emission computed tomography (SPECT) image features for early Parkinson's disease diagnosis.
• To identify a minimal set of features for accurate PD detection.
• To assess the clinical applicability of proposed features.

Main Methods:

• Utilized a dataset of 646 subjects for Parkinson's disease diagnosis.
• Extracted and analyzed 23 SPECT image features.
• Employed supervised machine learning models for classification.
• Evaluated model performance using balanced classification accuracy on independent test data.

Main Results:

• Achieved 94% balanced classification accuracy using the full set of 23 SPECT features.
• Demonstrated that comparable accuracy could be obtained using only eight selected features.
• Confirmed that all utilized features are extractable with standard clinical software.

Conclusions:

• SPECT image features, particularly a subset of eight, are effective for the early diagnosis of Parkinson's disease.
• The proposed feature extraction method is clinically feasible and straightforward.
• Machine learning models utilizing these SPECT features can significantly aid in early PD detection.