Applying Automated MR-Based Diagnostic Methods to the Memory Clinic: A Prospective Study.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Automated pattern recognition in structural MRI aids dementia diagnosis but often relies on preselected samples.
• Real-world dementia clinic data presents a more heterogeneous and challenging population for diagnostic algorithms.

Purpose of the Study:

• To evaluate a linear support vector machine for differentiating Alzheimer's disease (AD), frontotemporal dementia (FTD), Lewy body dementia, and healthy aging using 3D-T1 weighted MRI.
• To predict progression to AD in individuals with mild cognitive impairment (MCI) and quantify white matter hyperintensities.
• To assess the feasibility of MRI-based pattern recognition in a routine dementia clinic sample.

Main Methods:

• A linear support vector machine was trained on external data and applied to MRI datasets from a dementia clinic.
• The model differentiated between AD, FTD, Lewy body dementia, and healthy controls.
• Progression from MCI to AD was predicted, and white matter hyperintensities were quantified from FLAIR images.

Main Results:

• Distinguishing healthy elderly from dementia patients achieved high accuracy (AUC=0.97).
• Multi-class differentiation accuracy for AD (AUC=0.76) and FTD (AUC=0.78) was lower in the clinic sample compared to training data.
• Prediction of MCI progression to AD showed moderate accuracy (AUC=0.73), comparable to clinician performance (AUC=0.81).

Conclusions:

• MRI-based pattern recognition is feasible in routine dementia clinic samples but requires methodological refinement for accurate multi-class differential diagnosis.
• Future research should focus on improving automated multi-class diagnostic capabilities and application in diverse clinical populations.
• The study highlights the challenges in predicting MCI conversion in heterogeneous cohorts.