Deep Learning-based Hierarchical Brain Segmentation with Preliminary Analysis of the Repeatability and...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• Accurate brain subregion volumetry is crucial for neurological research and clinical diagnosis.
• Existing segmentation tools like SPM and FreeSurfer have limitations in repeatability and reproducibility.
• Deep learning offers potential for improved automated brain segmentation.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based hierarchical brain segmentation (DLHBS) method.
• To assess the repeatability and reproducibility of DLHBS for brain subregion volume estimation.
• To compare DLHBS performance against established tools: Statistical Parametric Mapping (SPM) and FreeSurfer (FS).

Main Methods:

• Employed hierarchical segmentation with multiple deep learning models for efficient brain subregion segmentation.
• Trained models using a large dataset (496 subjects) of T1-weighted MR images (T1WI) and validated with manual corrections.
• Evaluated repeatability and reproducibility using scan-rescan data from 11 healthy subjects across three MRI scanners.

Main Results:

• DLHBS demonstrated superior repeatability compared to SPM and FS across all eight evaluated regions, including gray matter, white matter, and hippocampus.
• DLHBS showed higher reproducibility than SPM in six regions and than FS in five regions.
• No instances of lower repeatability or reproducibility were observed with DLHBS.

Conclusions:

• DLHBS significantly outperforms SPM and FS in both repeatability and reproducibility for brain subregion volumetry.
• The DLHBS method provides a more reliable and accurate approach for brain image analysis.
• This advancement has implications for more robust neurological research and diagnostics.