Effects of Path-Finding Algorithms on the Labeling of the Centerlines of Circle of Willis Arteries

Quantitative analysis of intracranial vessel segments typically requires the identification of the vessels' centerlines, and a path-finding algorithm can be used to automatically detect vessel segments' centerlines. This study compared the performance of path-finding algorithms for vessel labeling. Three-dimensional (3D) time-of-flight magnetic resonance angiography (MRA) images from the publicly available dataset were considered for this study. After manual annotations of the endpoints of each vessel segment, three path-finding methods were compared: (Method 1) depth-first search algorithm, (Method 2) Dijkstra's algorithm, and (Method 3) A* algorithm. The rate of correctly found paths was quantified and compared among the three methods in each segment of the circle of Willis arteries. In the analysis of 840 vessel segments, Method 2 showed the highest accuracy (97.1%) of correctly found paths, while Method 1 and 3 showed an accuracy of 83.5% and 96.1%, respectively. The AComm artery was highly inaccurately identified in Method 1, with an accuracy of 43.2%. Incorrect paths by Method 2 were noted in the R-ICA, L-ICA, and R-PCA-P1 segments. The Dijkstra and A* algorithms showed similar accuracy in path-finding, and they were comparable in the speed of path-finding in the circle of Willis arterial segments.

Improved Skip-Gram Based on Graph Structure Information

Applying the Skip-gram to graph representation learning has become a widely researched topic in recent years. Prior works usually focus on the migration application of the Skip-gram model, while Skip-gram in graph representation learning, initially applied to word embedding, is left insufficiently explored. To compensate for the shortcoming, we analyze the difference between word embedding and graph embedding and reveal the principle of graph representation learning through a case study to explain the essential idea of graph embedding intuitively. Through the case study and in-depth understanding of graph embeddings, we propose Graph Skip-gram, an extension of the Skip-gram model using graph structure information. Graph Skip-gram can be combined with a variety of algorithms for excellent adaptability. Inspired by word embeddings in natural language processing, we design a novel feature fusion algorithm to fuse node vectors based on node vector similarity. We fully articulate the ideas of our approach on a small network and provide extensive experimental comparisons, including multiple classification tasks and link prediction tasks, demonstrating that our proposed approach is more applicable to graph representation learning.

HDGFI: Hierarchical Dual-Level Graph Feature Interaction Model for Personalized Recommendation

Under the background of information overload, the recommendation system has attracted wide attention as one of the most important means for this problem. Feature interaction considers not only the impact of each feature but also the combination of two or more features, which has become an important research field in recommendation systems. There are two essential problems in current feature interaction research. One is that not all feature interactions can generate positive gains, and some may lead to an increase in noise. The other is that the process of feature interactions is implicit and uninterpretable. In this paper, a Hierarchical Dual-level Graph Feature Interaction (HDGFI) model is proposed to solve these problems in the recommendation system. The model regards features as nodes and edges as interactions between features in the graph structure. Interaction noise is filtered by beneficial interaction selection based on a hierarchical edge selection module. At the same time, the importance of interaction between nodes is modeled in two perspectives in order to learn the representation of feature nodes at a finer granularity. Experimental results show that the proposed HDGFI model has higher accuracy than the existing models.

Fast and Accurate Lane Detection via Graph Structure and Disentangled Representation Learning

It is desirable to maintain high accuracy and runtime efficiency at the same time in lane detection. However, due to the long and thin properties of lanes, extracting features with both strong discrimination and perception abilities needs a huge amount of calculation, which seriously slows down the running speed. Therefore, we design a more efficient way to extract the features of lanes, including two phases: (1) Local feature extraction, which sets a series of predefined anchor lines, and extracts the local features through their locations. (2) Global feature aggregation, which treats local features as the nodes of the graph, and builds a fully connected graph by adaptively learning the distance between nodes, the global feature can be aggregated through weighted summing finally. Another problem that limits the performance is the information loss in feature compression, mainly due to the huge dimensional gap, e.g., from 512 to 8. To handle this issue, we propose a feature compression module based on decoupling representation learning. This module can effectively learn the statistical information and spatial relationships between features. After that, redundancy is greatly reduced and more critical information is retained. Extensional experimental results show that our proposed method is both fast and accurate. On the Tusimple and CULane benchmarks, with a running speed of 248 FPS, F1 values of 96.81% and 75.49% were achieved, respectively.

Peel Learning for Pathway-Related Outcome Prediction

MOTIVATION

Traditional regression models are limited in outcome prediction due to their parametric nature. Current deep learning methods allow for various effects and interactions and have shown improved performance, but they typically need to be trained on a large amount of data to obtain reliable results. Gene expression studies often have small sample sizes but high dimensional correlated predictors so that traditional deep learning methods are not readily applicable.

RESULTS

In this paper, we proposed peel learning, a novel neural network that incorporates the prior relationship among genes. In each layer of learning, overall structure is peeled into multiple local substructures. Within the substructure, dependency among variables is reduced through linear projections. The overall structure is gradually simplified over layers and weight parameters are optimized through a revised backpropagation. We applied PL to a small lung transplantation study to predict recipients' post-surgery primary graft dysfunction using donors' gene expressions within several immunology pathways, where PL showed improved prediction accuracy compared to conventional penalized regression, classification trees, feed-forward neural network, and a neural network assuming prior network structure. Through simulation studies, we also demonstrated the advantage of adding specific structure among predictor variables in neural network, over no or uniform group structure, which is more favorable in smaller studies. The empirical evidence is consistent with our theoretical proof of improved upper bound of PL's complexity over ordinary neural networks.

AVAILABILITY AND IMPLEMENTATION

PL algorithm was implemented in Python and the open-source code and instruction will be available at https://github.com/Likelyt/Peel-Learning.

Non-invasive reconstruction of dynamic myocardial transmembrane potential with graph-based total variation constraints

Non-invasive reconstruction of electrophysiological activity in the heart is of great significance for clinical disease prevention and surgical treatment. The distribution of transmembrane potential (TMP) in three-dimensional myocardium can help us diagnose heart diseases such as myocardial ischemia and ectopic pacing. However, the problem of solving TMP is ill-posed, and appropriate constraints need to be added. The existing state-of-art method total variation minimisation only takes advantage of the local similarity in space, which has the problem of over-smoothing, and fails to take into account the relationship among frames in the dynamic TMP sequence. In this work, the authors introduce a novel regularisation method called graph-based total variation to make up for the above shortcomings. The graph structure takes the TMP value of a time sequence on each heart node as the criterion to establish the similarity relationship among the heart. Two sets of phantom experiments were set to verify the superiority of the proposed method over the traditional constraints: infarct scar reconstruction and activation wavefront reconstruction. In addition, experiments with ten real premature ventricular contractions patient data were used to demonstrate the accuracy of the authors’ method in clinical applications.

Forced Edges and Graph Structure

For a degree sequence, we define the set of edges that appear in every labeled realization of that sequence as forced, while the edges that appear in none are define as forbidden. We examine the structure of graphs in which the degree sequences contain either forced or forbidden edges. The results include the determination of the structure of the forced or forbidden edge sets, the relationship between the sizes of forced and forbidden sets for a sequence, and the structural consequences to their realizations. This includes showing that the diameter of every realization of a degree sequence containing forced or forbidden edges is no greater than 3, and that these graphs are maximally edge-connected.

Probabilistic Multi-Sensor Fusion Based Indoor Positioning System on a Mobile Device

Nowadays, smart mobile devices include more and more sensors on board, such as motion sensors (accelerometer, gyroscope, magnetometer), wireless signal strength indicators (WiFi, Bluetooth, Zigbee), and visual sensors (LiDAR, camera). People have developed various indoor positioning techniques based on these sensors. In this paper, the probabilistic fusion of multiple sensors is investigated in a hidden Markov model (HMM) framework for mobile-device user-positioning. We propose a graph structure to store the model constructed by multiple sensors during the offline training phase, and a multimodal particle filter to seamlessly fuse the information during the online tracking phase. Based on our algorithm, we develop an indoor positioning system on the iOS platform. The experiments carried out in a typical indoor environment have shown promising results for our proposed algorithm and system design.