Automatic Craniomaxillofacial Landmark Digitization via Segmentation-Guided Partially-Joint Regression Forest Model and Multiscale Statistical Features

Abstract

Objective: The goal of this paper is to automatically digitize craniomaxillofacial (CMF) landmarks efficiently and accurately from cone-beam computed tomography (CBCT) images, by addressing the challenge caused by large morphological variations across patients and image artifacts of CBCT images.

Methods: We propose a segmentation-guided partially-joint regression forest (S-PRF) model to automatically digitize CMF landmarks. In this model, a regression voting strategy is first adopted to localize each landmark by aggregating evidences from context locations, thus potentially relieving the problem caused by image artifacts near the landmark. Second, CBCT image segmentation is utilized to remove uninformative voxels caused by morphological variations across patients. Third, a partially-joint model is further proposed to separately localize landmarks based on the coherence of landmark positions to improve the digitization reliability. In addition, we propose a fast vector quantization method to extract high-level multiscale statistical features to describe a voxel's appearance, which has low dimensionality, high efficiency, and is also invariant to the local inhomogeneity caused by artifacts.

Results: Mean digitization errors for 15 landmarks, in comparison to the ground truth, are all less than 2 mm.

Conclusion: Our model has addressed challenges of both interpatient morphological variations and imaging artifacts. Experiments on a CBCT dataset show that our approach achieves clinically acceptable accuracy for landmark digitalization.

Significance: Our automatic landmark digitization method can be used clinically to reduce the labor cost and also improve digitalization consistency.

Fig. 1

Inconsistent appearances of anatomical landmarks across different patients caused by (a) CMF deformity, and (b) metal effect. (c) Similar appearances in both MSCT and CBCT images.

Fig. 2

Flow chart of proposed landmark digitization method.

Fig. 3

Problems for the regression-forest-based landmark digitization. (a) Uninformative voxel in the mandible for localizing a landmark on upper tooth. (b) Incoherent displacements to the two same landmarks from two different patients.

Fig. 4

Schematic view of coherence of landmarks and multi-scale 3D patches for one voxel. (a) Definition of offset. (b) Stable substructures, where the substructure of Landmarks 1, 4, 5 and the substructure of landmarks 2, 3, 6 are relatively stable across different patients.

Fig. 5

Diagram of our fast VQ for the feature extraction.

Fig. 6

CMF landmarks annotated on a 3D skull model.

Fig. 7

Effect of weighted voting. The star is the digitization error without weighted voting.

Fig. 8

Landmark partition based on the landmark coherence. (a-b) Affinity matrices of landmarks from mandible and maxilla. (c) Partition result of landmarks. Note that the landmarks in the same group are shown in the same color.

Fig. 9

Quantitative comparisons. Note that the results of the same color across different subfigures are the same, and the number inside the parenthesis is the mean error for all the landmarks and patients. (a) Digitization errors with different segmentation accuracies. (b) Digitization errors of using partially-joint model or fully-joint model. (c) Digitization errors with and without MSCT for training. (d) Digitization errors of using Haar-like features or multi-scale statistical features. (e) Digitization errors of using different VQ methods. (f) Comparison with multi-atlas-based method.