Robust Registration of Statistical Shape Models for Unsupervised Pathology Annotation

We present a method to automatically label pathologies in volumetric medical data. Our solution makes use of a healthy statistical shape model to label pathologies in novel targets during model fitting. We achieve this using an EM algorithm: the E-step classifies surface points into pathological or healthy classes based on outliers in predicted correspondences, while the M-step performs probabilistic fitting of the statistical shape model to the healthy region. Our method is independent of pathology type or target anatomy, and can therefore be used for labeling different types of data. The method is able to detect pathologies with higher accuracy than standard robust detection algorithms, which we show using true positive rate and F1 scores. Furthermore, the method provides an estimate of the uncertainty of the synthesized label. The detection also directly improves surface reconstruction results, as shown by a decrease in the average and Hausdorff distances to ground truth. The method can be used for automated diagnosis or as a pre-processing step to accurately label large amounts of images.

[1]  George Lee,et al.  Image analysis and machine learning in digital pathology: Challenges and opportunities , 2016, Medical Image Anal..

[2]  Hans-Peter Meinzer,et al.  Statistical shape models for 3D medical image segmentation: A review , 2009, Medical Image Anal..

[3]  Andriy Myronenko,et al.  Point Set Registration: Coherent Point Drift , 2009, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[4]  Pavel Krsek,et al.  Robust Euclidean alignment of 3D point sets: the trimmed iterative closest point algorithm , 2005, Image Vis. Comput..

[5]  Tinashe Mutsvangwa,et al.  Construction and validation of image-based statistical shape and intensity models of bone , 2018, 2018 3rd Biennial South African Biomedical Engineering Conference (SAIBMEC).

[6]  Azriel Rosenfeld,et al.  Robust regression methods for computer vision: A review , 1991, International Journal of Computer Vision.

[7]  Thomas Vetter,et al.  Building Shape Models from Lousy Data , 2009, MICCAI.

[8]  F. E. Grubbs.  Procedures for Detecting Outlying Observations in Samples , 1969 .

[9]  Luc Van Gool,et al.  Matching of 3-D curves using semi-differential invariants , 1995, Proceedings of IEEE International Conference on Computer Vision.

[10]  Bernhard Egger,et al.  Occlusion-Aware 3D Morphable Models and an Illumination Prior for Face Image Analysis , 2018, International Journal of Computer Vision.

[11]  Thomas Gerig,et al.  Gaussian Process Morphable Models , 2016, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[12]  Matthew Toews,et al.  A Statistical Parts-Based Model of Anatomical Variability , 2007, IEEE Transactions on Medical Imaging.

[13]  Leonidas J. Guibas,et al.  Robust global registration , 2005, SGP '05.

[14]  R. Woods,et al.  Mathematical/computational challenges in creating deformable and probabilistic atlases of the human brain , 2000, Human brain mapping.

[15]  Yoshinobu Sato,et al.  Automated CT Segmentation of Diseased Hip Using Hierarchical and Conditional Statistical Shape Models , 2013, MICCAI.

[16]  Georg Langs,et al.  Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery , 2017, IPMI.

[17]  Thomas Vetter,et al.  Probabilistic Fitting of Active Shape Models , 2018, ShapeMI@MICCAI.

[18]  Anand Rangarajan,et al.  A new point matching algorithm for non-rigid registration , 2003, Comput. Vis. Image Underst..

[19]  John S. Duncan,et al.  Non-rigid Registration with Missing Correspondences in Preoperative and Postresection Brain Images , 2010, MICCAI.

[20]  T. Vetter,et al.  Shape Modeling Using Gaussian Process Morphable Models , 2017 .

[21]  Pascal A. Dufour,et al.  Pathology Hinting as the Combination of Automatic Segmentation with a Statistical Shape Model , 2012, MICCAI.