Validation of bone segmentation and improved 3-D registration using contour coherency in CT data

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of medical images, such as computed tomography (CT) scans, is crucial for reconstructing 3D models.
• Existing automatic segmentation techniques can produce errors, impacting the accuracy of subsequent 3D surface registration.
• Validating segmentation quality is essential for reliable downstream analysis.

Purpose of the Study:

• To present a novel method for validating computed tomography (CT) image segmentation.
• To improve the accuracy and efficiency of 3D surface registration from segmented CT data.
• To compare the performance of the validation method against human operators.

Main Methods:

• Segmentation validation by comparing contour shapes from neighboring CT slices.
• Parameterization of bone contours using normalized arc length and inscribed angle.
• Representation of contours as vectors in a K-dimensional space using Fourier Descriptors.
• Measurement of contour similarity (coherency) by comparing statistical properties of vector representations.

Main Results:

• The method effectively identifies low-coherency segmentations, indicating potential errors.
• Demonstrated high detection rates for both low-coherency (80-87.5%) and high-coherency (95.5-97.7%) segmentations compared to human operators.
• Removal of detected low-coherency segmentations significantly improved 3D bone surface model registration.
• Registration error reduced by over 500% and 280%, with computational performance improved by 540% and 791% for two segmentation methods.

Conclusions:

• The proposed contour-based validation method is highly effective for CT image segmentation.
• This technique enhances the reliability and efficiency of 3D surface reconstruction and registration.
• The method offers a significant improvement over existing automatic segmentation validation approaches.