Cephalometric analysis relies on accurate detection of craniomaxillofacial (CMF) landmarks from cone-beam computed tomography (CBCT) images. However, due to the complexity of CMF bony structures, it is difficult to localize landmarks efficiently and accurately. In this paper, we propose a deep learning framework to tackle this challenge by jointly digitalizing 105 CMF landmarks on CBCT images. By explicitly learning the local geometrical relationships between the landmarks, our approach extends Mask R-CNN for end-to-end prediction of landmark locations. Specifically, we first apply a detection network on a down-sampled 3D image to leverage global contextual information to predict the approximate locations of the landmarks. We subsequently leverage local information provided by higher-resolution image patches to refine the landmark locations. On patients with varying non-syndromic jaw deformities, our method achieves an average detection accuracy of 1.38± 0.95mm, outperforming a related state-of-the-art method.

Orthognathic surgical outcomes rely heavily on the quality of surgical planning. Automatic estimation of a reference facial bone shape significantly reduces experience-dependent variability and improves planning accuracy and efficiency. We propose an end-to-end deep learning framework to estimate patient-specific reference bony shape models for patients with orthognathic deformities. Specifically, we apply a point-cloud network to learn a vertex-wise deformation field from a patient's deformed bony shape, represented as a point cloud. The estimated deformation field is then used to correct the deformed bony shape to output a patient-specific reference bony surface model. To train our network effectively, we introduce a simulation strategy to synthesize deformed bones from any given normal bone, producing a relatively large and diverse dataset of shapes for training. Our method was evaluated using both synthetic and real patient data. Experimental results show that our framework estimates realistic reference bony shape models for patients with varying deformities. The performance of our method is consistently better than an existing method and several deep point-cloud networks. Our end-to-end estimation framework based on geometric deep learning shows great potential for improving clinical workflows.

In this paper, we present an efficient framework for 3D indoor scene labeling based on a Conditional Random Field model. To make this framework invariant to camera rotation, a novel feature vector is developed, which is simple but discriminative. Meanwhile, we re-define the pairwise potential to improve the performance. A method for learning the labeling compatibility is proposed to exploit the strong contextual relations between class labels. We evaluate our approach on three datasets and the experimental results show that it achieves higher accuracy compared with several state of the art researches.

In this paper, we propose an innovative algorithm for detecting single or multiple ground planes and estimating the orientation of the ground plane with a RGB-D camera. In our algorithm, we use Kernel Density Estimator for selecting the ground plane reliably under a given tilt angle. Camera tilt angle is estimated with probability theory and then used as a feedback to improve the accuracy of detection. Our algorithm is evaluated on several challenging data set and is shown to hold great superiority and effectiveness.

In this paper, we describe a method for analysis and identification of Electroencephalography (EEG) about visual stimulation. Here, ODDBALL task has been performed by using 4 different categories images to measure and analysis the EEG Signals, after which the P300 information of the visual stimulation can be detected and identified. In order to improve the identification ratio while avoiding the effects of noise, 1) in the pre-processing stage, we perform Normalization, Gaussian filter, and Non-maximum Suppression for emphasizing the patterns around the peak of the P300; 2) we construct high-dimensional vector which consists of the data obtained from 4 different electrodes. Experiments for comparing the method proposed with others using Linear discriminant analysis (LDA), K-nearest neighbour (k-NN) and Nearest mean (NM) have been implemented, the results of which has confirmed that our method owns an improvement of identification ratio.