3D statistical shape models incorporating 3D random forest regression voting for robust CT liver segmentation

Abstract

During image segmentation, 3D Statistical Shape Models (SSM) usually conduct a limited search for target landmarks within one-dimensional search profiles perpendicular to the model surface. In addition, landmark appearance is modeled only locally based on linear profiles and weak learners, altogether leading to segmentation errors from landmark ambiguities and limited search coverage. We present a new method for 3D SSM segmentation based on 3D Random Forest Regression Voting. For each surface landmark, a Random Regression Forest is trained that learns a 3D spatial displacement function between the according reference landmark and a set of surrounding sample points, based on an inifinite set of non-local randomized 3D Haar-like features. Landmark search is then conducted omni-directionally within 3D search spaces, where voxelwise forest predictions on landmark position contribute to a common voting map which reflects the overall position estimate. Segmentation experiments were conducted on a set of 45 CT volumes of the human liver, of which 40 images were randomly chosen for training and 5 for testing. Without parameter optimization, using a simple candidate selection and a single resolution approach, excellent results were achieved, while faster convergence and better concavity segmentation were observed, altogether underlining the potential of our approach in terms of increased robustness from distinct landmark detection and from better search coverage.

Automatic segmentation of the thumb trapeziometacarpal joint using parametric statistical shape modelling and random forest regression voting

“…The mean error in the first metacarpal was 0.997 mm ± 0.372 mm, and 0.564 mm ± 0.282 mm in the trapezium, which may be considered reasonable depending on the purpose of the model. The mean % volume overlap was 84.12 % in the first metacarpal and 86.03 % in the trapezium, is comparable to the intermediate rigid model fitting results on the liver reported by Norajitra, Meinzer and Maier-Hein (2015), except in a much smaller joint. However, for the purpose of contact biomechanics at the articular surfaces of the joint, these errors may require further reduction, as the maximum errors (2.110 mm in the first metacarpal and 1.776 mm in the trapezium) are comparable to the size of the joint space in the TMC joint (~ 2 mm).…”

“…3D Haar-like features) to train a forest of decision trees to predict the most likely image location of the desired model. This has been demonstrated in 2D with facial recognition (Cootes et al 2012) and 2D segmentation of the proximal femur (Lindner et al 2013), and more recently demonstrated in 3D in the Liver (Norajitra et al 2015). This method can be combined with parametric statistical shape modelling to create an automatic segmentation pipeline that has increased robustness to initialisation, increased speed of segmentation, and automatic meshing for downstream analysis.…”

“…Further improvements in segmentation accuracy could be achieved by relaxing the SSM constraints in the final iteration of fitting. In another study, Norajitra, Meinzer and Maier-Hein (2015) reported an improvement in segmentation accuracy with the application of a deformable surface algorithm during the final step in fitting. This was to allow the surface to deviate from the rigid shape constraint of the SSM, and to compensate for the liver’s high variability in shape.…”

“…Haar-like features are used in image object recognition and have been used in image segmentation in 2D (Cootes, Ionita, Lindner and Sauer 2012) and 3D (Norajitra and Maier-Hein 2017, Norajitra, Meinzer and Maier-Hein 2015). Haar-like features are calculated by comparing the difference in summed pixel intensity between regions of pixels (in 2D) or voxels (in 3D) within a bounding box These regions may be labelled as ‘dark’ and ‘light’, where the bounds of the ‘light’ region can be randomised to create an infinite set of 3D Haar-like features (Lindner, Thiagarajah, Wilkinson, Consortium, Wallis and Cootes 2013, Norajitra and Maier-Hein 2017).…”

MRI-Derived Dural Sac and Lumbar Vertebrae 3D Volumetry Has Potential for Detection of Marfan Syndrome

“…Despite the development of various automated segmentation methods, including heuristically customized appearance models and automatic selection of optimal local features [41], these methods are limited by their reliance on local image information and the use of heuristic or weak learning methods [40,42]. However, models utilizing randomized regression forests have emerged, offering robust learning that consider non-local image information and omni-directional landmark detection, thereby enhancing segmentation accuracy and robustness across various organs without the need for prior model initialization [40,[43][44][45].…”