Automatically computed rating scales from MRI for patients with cognitive disorders

Area of Science:

• Neuroimaging
• Computational Medicine
• Cognitive Neurology

Background:

• Visual Magnetic Resonance Imaging (MRI) rating scales are crucial for diagnosing cognitive disorders but can be subjective.
• Developing objective, computational methods for these scales could enhance diagnostic reliability and consistency.
• Medial temporal lobe atrophy (MTA), global cortical atrophy (GCA), and white matter hyperintensities (WMHs) are key imaging biomarkers.

Purpose of the Study:

• To determine if visual MRI rating scales for cognitive disorders can be accurately estimated computationally.
• To compare the diagnostic performance of visually rated scales versus computationally derived scales in differential diagnostics.

Main Methods:

• Extracted imaging biomarkers (volumetry, voxel-based morphometry) from T1-weighted and FLAIR MRI scans.
• Developed a regression model to estimate visual rating scale values (MTA, GCA, Fazekas scale for WMHs) from biomarkers.
• Validated models using the Amsterdam Dementia Cohort (ADC), PredictND, and Alzheimer's Disease Neuroimaging Initiative (ADNI) cohorts.

Main Results:

• High correlations between visual and computed scales: 0.83/0.78 for MTA, 0.64/0.64 for GCA, and 0.76/0.75 for Fazekas (ADC/PredictND).
• Computed GCA achieved the highest accuracy (0.75-0.86) in differentiating dementias from cognitively normal subjects.
• Computed scales showed statistically significant higher balanced accuracies than visual scales for MTA and GCA.

Conclusions:

• Automated estimation of MTA, GCA, and WMHs is reliable, providing consistent imaging biomarkers for cognitive disorder diagnosis.
• Computational scales offer high diagnostic accuracy, potentially aiding clinicians of all experience levels.
• This automated approach enhances the objectivity and reproducibility of MRI-based assessments in dementia diagnostics.