A semi-supervised learning approach for automated 3D cephalometric landmark identification using computed tomography

Abstract

Identification of 3D cephalometric landmarks that serve as proxy to the shape of human skull is the fundamental step in cephalometric analysis. Since manual landmarking from 3D computed tomography (CT) images is a cumbersome task even for the trained experts, automatic 3D landmark detection system is in a great need. Recently, automatic landmarking of 2D cephalograms using deep learning (DL) has achieved great success, but 3D landmarking for more than 80 landmarks has not yet reached a satisfactory level, because of the factors hindering machine learning such as the high dimensionality of the input data and limited amount of training data due to the ethical restrictions on the use of medical data. This paper presents a semi-supervised DL method for 3D landmarking that takes advantage of anonymized landmark dataset with paired CT data being removed. The proposed method first detects a small number of easy-to-find reference landmarks, then uses them to provide a rough estimation of the all landmarks by utilizing the low dimensional representation learned by variational autoencoder (VAE). The anonymized landmark dataset is used for training the VAE. Finally, coarse-to-fine detection is applied to the small bounding box provided by rough estimation, using separate strategies suitable for the mandible and the cranium. For mandibular landmarks, patch-based 3D CNN is applied to the segmented image of the mandible (separated from the maxilla), in order to capture 3D morphological features of mandible associated with the landmarks. We detect 6 landmarks around the condyle all at once rather than one by one, because they are closely related to each other. For cranial landmarks, we again use the VAE-based latent representation for more accurate annotation. In our experiment, the proposed method achieved a mean detection error of 2.88 mm for 90 landmarks using only 15 paired training data.

Accuracy of automated 3D cephalometric landmarks by deep learning algorithms: systematic review and meta-analysis

“…Six studies used an image dataset of CT, two studies CBCT and four studies used both. One study used as dataset both annotated CBCT images (labeled images) and not-annotated ones (unlabeled) [29] and three studies [30][31][32] used a dataset composed by labeled CT images and a landmark dataset of the 3D positions of landmarks from CT. One study reported the mean error in pixels dimension [33] instead of mm, and for this review, it was converted in mm using the pixel-to-mm conversion rate reported in the article. Studies detected a mean (± standard deviation) of 47(± 35) landmarks, with a 5-105 range.…”

“…For eleven studies the reported outcome was the mean error and the standard deviation between manual and automatic landmarking and they were included in the meta-analysis [29,30,[33][34][35][36][37][38][39][40][41]. Four studies didn't report the standard deviation [31,32,42,43]; thus, they were excluded. Detailed information about studies' characteristics can be found in Table 1.…”

Machine learning-based signal quality assessment for cardiac volume monitoring in electrical impedance tomography

“…The decoder D is trained to generate normal CVS-like output. In other words, operation D • E transforms X to lie in or near the learned manifold using normal CVS data [47,58]. Therefore, the residual r can be viewed as an anomaly score, where r is small if X is normal CVS data, and large if X is motion-influenced CVS data.…”

“…where KL is Kullback-Leibler divergence and Σ is a 10 × 10 diagonal matrix whose (i, i) entry is σ i . This term enables VAE to learn dense and smooth latent space embedding in or near N (0, I) [31,50,58].…”

Three-Dimensional Craniofacial Landmark Detection in Series of CT Slices Using Multi-Phased Regression Networks

“…A knowledge-based method [44] was reported in 2015. Various kinds of learning-based methods [45][46][47][48][49][50][51][52][53][54][55] have been reported. In our experience with 2D cephalograms [27][28][29], a multi-phased deep learning system was able to predict coordinate values with high precision.…”