Skip to main content

Table 1 Solutions for segmentation of CT images in previous research

From: Utility of a novel integrated deep convolutional neural network for the segmentation of hip joint from computed tomography images in the preoperative planning of total hip arthroplasty

  Year of publication Title of article Contribution Advantage Inconvenient
1 2017 Accurate pelvis and femur segmentation in Hip CT with a novel patch-based refinement They have presented a coarse-to-fine hip CT segmentation framework that consisted of region growing-based preprocessing, CRF-based coarse segmentation and patch-based refinement. The experimental results on 60 CT hips (120 hemi-hips) demonstrate the effectiveness of their method (1) It starts with coarse segmentation techniques for obtaining the bone boundaries, followed by the bone refinement using a patch-based algorithm that is navigated by the extracted bone boundaries
(2) GLCM is first used for bone classification and its effectiveness is demonstrated
(3) the existing methods perform the label propagation for all voxels of the image to be segmented
(1) The total hip must be included in each volume. The position normalization for VOI is based on whole hip, so it will be a wrong indication in partial hip CT data
(2) The highly refined manual segmentation is required. For each training sample, the radiology experts put a significant amount of effort into completing it
(3) The method needs a long time computation
2 2018 Automated muscle segmentation from CT images of the hip and thigh using a hierarchical multi-atlas method In this paper, they proposed a hierarchization of the multi-atlas method to reduce the inter-patient variability in muscles. Intermediate segmentation results of more easily segmentable structures, that is, skin, bones, and entire muscle, were effectively utilized for spatial normalization to reduce inter-patient variabilities of the final target structures of individual muscles Significant improvement was observed by using the proposed hierarchical strategy. Although the individual muscles have large inter-patient variabilities, the spatial normalization using the region pre-segmented in the previous stage reduces the inter-patient variability. Significant improvements in accuracy were observed among all individual muscles around the thigh segment A limitation of the proposed method, especially in its application in biomechanical simulation, is the lack of imaging of tendons, ligaments, and attachment regions
3 2017 Fully automated segmentation of a hip joint using the patient-specific
optimal thresholding and watershed algorithm
This study proposed a fully automated segmentation method for a hip joint using the complementary characteristics between the patient-specific optimal thresholding and the watershed algorithm The thresholding method generates patches which are often not closed but contain regional information; and the watershed algorithm generates patches which always have closed boundaries but have no regional information
Clinical case studies with eight sets of CT scan data demonstrated that the proposed method can reliably segment a hip joint with high speed and accuracy without the aid of a prerequisite dataset and user manual intervention
(1) The proposed method was validated only with eight cases. (2) The accuracy of the proposed method is affected by the closed patches of the watershed algorithm
(3) A use of primitive spheres in the proposed method may not be effective in the CT scan data where the femoral head is severely deformed due to diseases such as avascular necrosis