Tumor boundary detection in ultrasound imagery using multi-scale generalized gradient vector flow

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Computational Anatomy

Background:

• Precise tumor boundary detection is crucial for effective high-intensity focused ultrasound (HIFU) ablation.
• Accurate segmentation in ultrasound images helps prevent tumor recurrence.
• Existing methods face challenges with spurious boundary attenuation.

Purpose of the Study:

• To introduce a novel deformable snake model, the multi-scale generalized gradient vector flow (MS-GGVF), for ultrasound image segmentation.
• To enhance tumor boundary detection in HIFU ablation systems.
• To address limitations in current segmentation techniques.

Main Methods:

• Developed the MS-GGVF model by assigning Gaussian filter standard deviation as scales.
• Utilized a signed distance map for gradient direction and magnitude to refine edge maps.
• Implemented a fast generalized gradient vector flow computation using an augmented Lagrangian method.

Main Results:

• Experimental segmentations closely matched physician-delineated ground truths.
• Achieved high area overlap measures between segmented and true boundaries.
• Demonstrated low mean contour distance, indicating precise boundary localization.

Conclusions:

• The proposed MS-GGVF algorithm offers robust and reliable tumor boundary detection.
• The method is precise for ultrasound image segmentation in HIFU ablation.
• This technique shows significant potential for improving HIFU ablation outcomes.