Texture-based, automatic contour validation for online adaptive replanning: A feasibility study on abdominal organs

Area of Science:

• Medical Imaging
• Radiation Oncology
• Computational Pathology

Background:

• Manual contour evaluation is a bottleneck in radiation therapy planning, hindering online adaptive replanning (OLAR).
• Automating contour quality assessment is crucial for efficient OLAR workflows.

Purpose of the Study:

• To develop an automatic approach for rapid contour quality evaluation using image texture features.
• To facilitate the routine practice of online adaptive replanning (OLAR) by reducing manual interaction.

Main Methods:

• A decision tree model was constructed using texture features from core, inner, and outer subregions of pancreas head and duodenum contours.
• Texture features, including principal component analysis and GLCM-based cluster prominence, were extracted from 275 CT image sets.
• The model was trained and validated using both ground-truth and inaccurately propagated contours.

Main Results:

• The model achieved high accuracy in identifying accurate vs. inaccurate contours: 85%/91% sensitivity/specificity for pancreas head and 92%/92% for duodenum.
• Low false-positive rates of 9% (pancreas head) and 8% (duodenum) were observed.
• Contour evaluation was rapid, with execution times under 15 seconds on a standard desktop computer.

Conclusions:

• Quantitative image texture features can effectively automate contour quality validation in radiation therapy.
• The developed model accurately classifies contours, potentially integrating into automated pipelines for auto-segmentation and correction.
• This automation can significantly streamline OLAR, replacing time-consuming manual processes.