Edge-Aware Diffusion Segmentation Model With Hessian Priors for Automated Diaphragm Thickness Measurement in...

Area of Science:

• Medical imaging analysis
• Biomedical engineering
• Computational anatomy

Background:

• Diaphragm thickness is a key indicator for respiratory function assessment.
• Manual ultrasound-based measurement is subjective, slow, and prone to noise.
• Accurate and reliable diaphragm thickness quantification is clinically needed.

Purpose of the Study:

• To develop an automated, accurate, and reliable method for diaphragm thickness measurement using ultrasound.
• To improve fascia segmentation in ultrasound images for better biometric analysis.
• To address limitations of manual measurement, including subjectivity and noise sensitivity.

Main Methods:

• Introduced an edge-aware diffusion segmentation model (ESADiff) incorporating structural fascia knowledge.
• Employed a diffusion model with iterative denoising for edge-preserving feature learning.
• Developed an anisotropic edge-sensitive annotation refinement module and curvature-aware deformable convolutions with an edge-prior ranking loss function.

Main Results:

• ESADiff achieved finer fascia segmentation compared to state-of-the-art techniques.
• Significantly improved the accuracy and reliability of diaphragm thickness measurements.
• Demonstrated robust generalization across diaphragm, calf, and tongue muscle ultrasound datasets.

Conclusions:

• The proposed ESADiff model offers a superior approach for diaphragm thickness measurement from ultrasound.
• The method effectively leverages prior structural and shape knowledge for enhanced segmentation.
• ESADiff shows significant potential for improving clinical assessment of respiratory function.