How segmentation methods affect hippocampal radiomic feature accuracy in Alzheimer's disease analysis?

Area of Science:

• Neuroimaging
• Radiomics
• Biomarkers

Background:

• Hippocampal radiomic features (HRFs) are potential biomarkers for Alzheimer's disease (AD).
• The impact of various hippocampal segmentation methods on HRFs in AD analysis remains unclear.
• This study investigates how different segmentation techniques influence HRF accuracy for AD detection.

Purpose of the Study:

• To evaluate the consistency of HRFs across seven different hippocampal segmentation methods.
• To assess the performance of machine learning models in classifying AD vs. normal control (NC) using HRFs derived from various segmentation techniques.

Main Methods:

• 1650 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were analyzed.
• HRFs (intensity, shape, texture) were extracted from structural MRI.
• Consistency was validated using the European DTI Study on Dementia (EDSD) cohort.

Main Results:

• HRFs demonstrated high measurement consistency (R > 0.7) across segmentations.
• Significant consistency was observed between normal control (NC), mild cognitive impairment (MCI), and AD groups (R > 0.8).
• Naïve majority voting segmentation yielded the best NC vs. AD classification performance.

Conclusions:

• HRFs are consistent across diverse hippocampal segmentation methods.
• The naïve majority voting method, with sufficient segmentation, offers optimal performance for AD classification.
• These findings support the utility of HRFs in AD diagnosis and research.