Design and validation of a semi-automatic bone segmentation algorithm from MRI to improve research efficiency

Abstract

Segmentation of medical images into different tissue types is essential for many advancements in orthopaedic research; however, manual segmentation techniques can be time- and cost-prohibitive. The purpose of this work was to develop a semi-automatic segmentation algorithm that leverages gradients in spatial intensity to isolate the patella bone from magnetic resonance (MR) images of the knee that does not require a training set. The developed algorithm was validated in a sample of four human participants (in vivo) and three porcine stifle joints (ex vivo) using both magnetic resonance imaging (MRI) and computed tomography (CT). We assessed the repeatability (expressed as mean ± standard deviation) of the semi-automatic segmentation technique on: (1) the same MRI scan twice (Dice similarity coefficient = 0.988 ± 0.002; surface distance = - 0.01 ± 0.001 mm), (2) the scan/re-scan repeatability of the segmentation technique (surface distance = - 0.02 ± 0.03 mm), (3) how the semi-automatic segmentation technique compared to manual MRI segmentation (surface distance = - 0.02 ± 0.08 mm), and (4) how the semi-automatic segmentation technique compared when applied to both MRI and CT images of the same specimens (surface distance = - 0.02 ± 0.06 mm). Mean surface distances perpendicular to the cartilage surface were computed between pairs of patellar bone models. Critically, the semi-automatic segmentation algorithm developed in this work reduced segmentation time by approximately 75%. This method is promising for improving research throughput and potentially for use in generating training data for deep learning algorithms.

Figure 1

Overview of comparisons made between segmentation techniques. In human participants, the semi-automatic segmentations of T1 VIBE MRI scans were compared to: (1) repeated semi-automatic segmentations of the same T1 VIBE MRI scans, (2) semi-automatic segmentations of second T1 VIBE MRI scans of the same participants, and (3) manual segmentations of DESS MRI scans of the same participants. In porcine specimens, the semi-automatic segmentations of T1 VIBE MRI scans were compared to the semi-automatic segmentations of CT scans of the same specimens. T1 VIBE T1-weighted Volume-Interpolated Breathhold Examination with Water Excitation, DESS Double Echo Steady-State, MRI Magnetic Resonance Imaging, CT Computed Tomography.

Figure 2

Sagittal (A) T1 VIBE and (B) DESS MR images were acquired of each human participant’s dominant knee. While DESS images have been reliably used for knee joint bone and cartilage model generation in the past, T1 VIBE had greater contrast between the bone and adjacent cartilage (0.95 vs. 0.72). DESS Double Echo Steady-State, T1 VIBE T1-weighted Volume-Interpolated Breathhold Examination with Water Excitation.

Figure 3

Imaging protocols for the in vivo and ex vivo arms of the experiment. (A) T1 VIBE and DESS MR images were acquired of each human (in vivo) participant’s dominant knee. (B) The porcine (ex vivo) imaging protocol consisted of CT imaging followed by the same T1 VIBE MRI sequence performed on the human participants. T1 VIBE T1-weighted Volume-Interpolated Breathhold Examination with Water Excitation, DESS Double Echo Steady-State, MR magnetic resonance, CT computed tomography.

Figure 4

Semi-automatic segmentation steps for isolating bone from either MR or CT images. (1) A Canny edge detection filter is used to identify all edges in each image. The edges are overlaid in yellow on each image. (2) The image volume is cropped around the bone of interest (patella). (3) On a single starting image slice, the user selects the edges of the patella previously identified by the Canny filter. The selected edges are highlighted in cyan. (4) This starting slice is then used to determine the edges nearest these points in adjacent slices in the image volume. The user can remove stray edge points in any slice before proceeding. (5) The identified bone edges are converted into a three-dimensional point cloud.

Figure 5

(A) Lateral view of a 3D point cloud model generated from semi-automatic segmentations of a human patella. (B) Lateral view of a 3D surface mesh generated from the 3D point cloud in (A). (C) Posterior view of the 3D surface mesh in (B). The gray region depicts the area in which cartilage is located. A anterior, P posterior, S superior, I inferior, M medial, L lateral).

Figure 6

Posterior views of representative 3D bone models of a human patella generated using (A) semi-automatic segmentation of sagittal T1 VIBE MR images and (B) manual segmentation of sagittal DESS MR images. The same post-processing and smoothing operators were applied to both models. The DESS images had a larger slice thickness as compared to the T1 VIBE images (1 mm vs. 0.7 mm), which may contribute to differences in smoothness between the models. (C) The surface distances map shows strong agreement between the 3D models. Blue indicates regions where the semi-automatic segmentation model was larger than the manual segmentation model, whereas red indicates regions where the manual segmentation model was larger than the semi-automatic segmentation model. The black dashed line represents the bone region analyzed after the 25% perimeter reduction of the cartilage boundary, which was implemented to minimize edge effects. Surface distance was defined as the difference between the x-coordinates of the two bone models. T1 VIBE T1-weighted Volume-Interpolated Breathhold Examination with Water Excitation, DESS Double Echo Steady-State, MR magnetic resonance, S superior, I inferior, M medial, L lateral).