Image segmentation by EM-based adaptive pulse coupled neural networks in brain magnetic resonance imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Accurate segmentation of brain tissues in Magnetic Resonance Imaging (MRI) is crucial for medical research and clinical applications.
• Existing segmentation methods face challenges in achieving optimal precision, particularly for gray matter (GM) and brain parenchyma (GM+WM).

Purpose of the Study:

• To develop and evaluate an automatic hybrid image segmentation model for brain MRI.
• To enhance segmentation accuracy by integrating the Expectation Maximization (EM) model with a spatial Pulse Coupled Neural Network (PCNN) using an adaptive parameter tuning mechanism.

Main Methods:

• A novel adaptive EM-PCNN model was proposed, integrating EM for segmentation evaluation and PCNN parameter adaptation.
• The model was used to segment brain MR images into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).
• Performance was compared against non-adaptive EM-PCNN, EM, and Bias Corrected Fuzzy C-Means (BCFCM) algorithms using golden standard comparisons.

Main Results:

• The adaptive EM-PCNN demonstrated significantly superior performance in gray matter segmentation compared to non-adaptive EM-PCNN, EM, and BCFCM.
• For brain parenchyma segmentation, the adaptive EM-PCNN significantly outperformed BCFCM and showed better average performance than non-adaptive EM-PCNN and EM.
• The model generated accurate boundaries for GM and brain parenchyma (GM+WM).

Conclusions:

• The adaptive EM-PCNN model provides superior results for segmenting gray matter and brain parenchyma in brain MRI.
• The integration of adaptive parameter tuning significantly improves segmentation accuracy over non-adaptive approaches.
• This hybrid model represents a promising advancement in automated brain MRI segmentation techniques.