Multi-phase simultaneous segmentation of tumor in lung 4D-CT data with context information

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Diagnosis

Background:

• Lung 4D computed tomography (4D-CT) is vital for precise radiotherapy by characterizing respiratory motion for accurate target definition.
• Manual segmentation of lung tumors in 4D-CT data is labor-intensive due to the large number of slices.
• Accurate tumor segmentation remains a significant challenge in computer-aided diagnosis.

Purpose of the Study:

• To develop a convenient and robust method for lung 4D-CT tumor segmentation.
• To improve the accuracy and efficiency of tumor delineation in radiotherapy planning.
• To address the challenges of manual segmentation workload and segmentation difficulties.

Main Methods:

• Proposed an improved graph cut algorithm incorporating context information constraints.
• Combined all respiratory phases of lung 4D-CT into a global graph with a global energy function.
• Enforced context cost terms between neighboring phases for robust segmentation across all phases via max-flow/min-cut optimization.

Main Results:

• Achieved simultaneous and robust segmentation of lung tumors across all 4D-CT phases.
• Demonstrated superior accuracy and robustness compared to existing methods like graph cut without context, level set, and graph cut with star shape prior.
• Validated effectiveness through experiments on 10 diverse lung 4D-CT cases.

Conclusions:

• The proposed improved graph cut algorithm offers a more accurate and robust solution for lung 4D-CT tumor segmentation.
• This method effectively reduces the manual workload for clinicians and improves computer-aided diagnosis.
• The context information constraint is key to achieving reliable segmentation in dynamic imaging like 4D-CT.