Tracking fuzzy borders using geodesic curves with application to liver segmentation on planning CT

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiotherapy Planning

Background:

• Automatic liver segmentation on noncontrast-enhanced planning CT images is challenging due to fuzzy borders.
• Oversegmentation often occurs at the liver-chestwall and liver-heart interfaces.

Purpose of the Study:

• To develop a robust method for tracking fuzzy borders between the liver and adjacent organs.
• To apply this method for improved automatic liver segmentation in noncontrast-enhanced planning CT images.

Main Methods:

• Fuzzy borders were identified by minimizing gradient-weighted path length using the fast-marching method.
• The method incorporated fuzzy border tracking into an automatic segmentation scheme using adaptive thresholding and a geodesic active contour model.
• Validation was performed on planning CT images from 15 liver cancer patients.

Main Results:

• The proposed method achieved an average Dice similarity coefficient of 0.930 ± 0.015, outperforming segmentation without fuzzy border tracking (0.912 ± 0.020).
• Liver volumes generated by the method showed excellent agreement with manual delineations (correlation coefficient 0.98).
• Fuzzy border tracking significantly improved segmentation performance and reduced contour leakage.

Conclusions:

• The developed method provides accurate liver segmentation on noncontrast-enhanced planning CT images.
• Tracking fuzzy borders effectively minimizes contour leakage during active contour evolution, enhancing segmentation reliability.