Removing outliers from the normative database improves regional atrophy detection in single-subject voxel-based...

Area of Science:

• Neuroimaging
• Radiology
• Medical image analysis

Background:

• Single-subject voxel-based morphometry (VBM) uses T1-weighted MRI and a normative database (NDB) to detect regional atrophy.
• Outliers within the NDB can reduce VBM's sensitivity in identifying neurodegenerative diseases.

Purpose of the Study:

• To introduce and evaluate a method for removing outliers from NDBs ('NDB cleaning').
• To assess the impact of NDB cleaning on VBM performance for Alzheimer's disease (AD) and frontotemporal lobar degeneration (FTLD) detection.

Main Methods:

• T1-weighted MRI scans from 81 patients (AD/FTLD) and 37 healthy controls were analyzed.
• Two NDBs were used: scanner-specific and multi-scanner.
• Three quality metrics were applied for outlier detection and removal from the NDB.

Main Results:

• NDB cleaning significantly increased VBM sensitivity for AD and FTLD detection, particularly with a multi-scanner NDB (0.47 to 0.61).
• Specificity remained at 100% across all tested settings.
• The improvements were statistically significant.

Conclusions:

• NDB cleaning is a viable method to enhance VBM sensitivity for detecting AD and FTLD.
• This approach improves diagnostic performance without compromising specificity or increasing false positives.