Binary classification of ¹⁸F-flutemetamol PET using machine learning: comparison with visual reads and structural MRI

Area of Science:

• Neuroimaging
• Machine Learning
• Biomarkers

Background:

• Amyloid imaging using (18)F-flutemetamol PET is crucial for diagnosing Alzheimer's disease (AD).
• Binary classification of amyloid scans as normal or 'Alzheimer-like' is clinically significant.
• Supervised machine learning, specifically support vector machines (SVM), can potentially automate and standardize image interpretation.

Purpose of the Study:

• To evaluate if SVM can replicate visual classification of (18)F-flutemetamol PET scans.
• To identify image components with the highest diagnostic value according to SVM.
• To compare SVM-based classification of amyloid scans with SVM-based classification of structural MRI data.

Main Methods:

• Support vector machines (SVM) with a linear kernel were applied to (18)F-flutemetamol PET and volumetric MRI scans from 72 subjects (27 AD, 20 MCI, 25 controls).
• A leave-one-out cross-validation approach was used to train and test the SVM classifiers.
• Classification performance was assessed by comparing SVM results to visual reads and clinical diagnoses.

Main Results:

• The SVM classifier accurately replicated visual read assignments for (18)F-flutemetamol scans with 100% accuracy.
• Key brain regions contributing to classification included the striatum, precuneus, cingulate, and middle frontal gyrus.
• The (18)F-flutemetamol-based SVM classifier showed higher specificity (92%) than the MRI-based classifier (68%) for distinguishing AD from controls, and better identified MCI converters.

Conclusions:

• SVM can reliably replicate visual classification of (18)F-flutemetamol amyloid PET scans.
• (18)F-flutemetamol PET combined with SVM offers high specificity for Alzheimer's disease detection.
• This approach shows promise for improved diagnostic accuracy, particularly in differentiating between MCI converters and non-converters.