Hippocampus Segmentation Based on Iterative Local Linear Mapping With Representative and Local Structure-Preserved...

Area of Science:

• Medical Imaging Analysis
• Neuroscience
• Machine Learning

Background:

• Accurate hippocampus segmentation is crucial for diagnosing neurological conditions like Alzheimer's disease.
• Patch-based multi-atlas segmentation (PBMAS) is a common method but suffers from high computational costs and reliance on registration accuracy.

Purpose of the Study:

• To develop a novel, registration-free method for accurate and robust hippocampus segmentation.
• To overcome the limitations of existing PBMAS approaches.

Main Methods:

• Proposed a novel method based on iterative local linear mapping (ILLM) with feature embedding.
• Utilized semi-supervised deep autoencoder (SSDA) for nonlinear transformation of MR patches.
• Employed local linear mapping and threshold segmentation for preliminary prediction, refined by ILLM.

Main Results:

• Achieved superior performance compared to state-of-the-art PBMAS and classification-based methods.
• Obtained mean Dice similarity coefficients of 0.8852±0.0203 (1.5T) and 0.8783 ± 0.0251 (3.0T) for bilateral hippocampus segmentation.
• Demonstrated accurate and robust segmentation without the need for registration.

Conclusions:

• The proposed ILLM method provides an accurate and efficient alternative for hippocampus segmentation.
• This registration-free approach has significant implications for the diagnosis of neurological diseases.
• The method shows strong potential for clinical application in medical image analysis.