Interactive medical image segmentation using PDE control of active contours.

Area of Science:

• Medical image analysis
• Computational anatomy
• Image segmentation

Background:

• Automated segmentation of injured or unusual anatomical structures in medical imagery remains a significant challenge.
• Existing methods often struggle to achieve the accuracy and consistency of human experts on real-world medical data.

Purpose of the Study:

• To develop an easy-to-use and consistent interactive segmentation technique for medical images.
• To transform the interactive segmentation problem into a control synthesis problem for improved performance.

Main Methods:

• A level set partial differential equation (PDE) was used as the basis for an open-loop system.
• User input and an observer-like system were employed to perturb the state and dynamics of the level set PDE, creating a closed-loop system.
• The input structure was designed for intuitive user control without requiring mathematical parameter knowledge.

Main Results:

• The closed-loop system demonstrated desirable behavior, leading to accurate segmentation of complex structures.
• The technique was successfully applied to segment a patellar tendon in magnetic resonance imaging (MRI) and a shattered femur in computed tomography (CT).

Conclusions:

• The proposed control synthesis approach offers an effective solution for interactive medical image segmentation.
• This method enhances segmentation consistency and accuracy, particularly for challenging anatomical cases, by integrating user expertise seamlessly.