Multiple Hidden Markov Model for Pathological Vessel Segmentation

Area of Science:

• Medical Imaging
• Computational Biology
• Image Analysis

Background:

• Accurate vessel boundary delineation is crucial for medical diagnosis but is hindered by pathologies causing obscure boundaries and vessel-like structures.
• Existing segmentation methods struggle with the complexity introduced by pathological conditions in vascular imaging.

Purpose of the Study:

• To develop a novel segmentation method that accurately delineates vessel boundaries, particularly in the presence of pathologies.
• To address the limitations of current techniques in segmenting complex and obscured vascular structures.

Main Methods:

• A novel segmentation method employing multiple Hidden Markov Models (HMMs) was developed.
• The method utilizes a vessel axis plus cross-section model to constrain the segmentation process, ensuring physiological accuracy and computational efficiency.
• It leverages redundant information from multiple vessel-specific features and the statistical properties of HMMs to handle complex pathological cases.

Main Results:

• The method achieved a 91% overlap ratio on synthetic complex-structured datasets.
• Validation on real clinical data, specifically pathological abdominal arteries, demonstrated promising performance.
• The proposed method outperformed two state-of-the-art methods on both synthetic and real clinical datasets.

Conclusions:

• The novel HMM-based segmentation method effectively delineates pathological vessels with high accuracy.
• The approach offers a physiologically accurate and computationally effective solution for challenging vascular segmentation tasks.
• This method shows significant potential for improving the analysis of vascular pathologies in medical imaging.