Patient-ventilator asynchrony (PVA) is a significant challenge in mechanical ventilation, affecting approximately 25% of intensive care unit patients and increasing the risk of lung and diaphragm injury. Segmenting breathing cycles from long ventilation waveforms is essential for the reliable detection of PVA events. However, existing segmentation methods present several limitations: manual annotation is time-consuming; fixed-length window and rule-based segmentation methods lack adaptability to varying respiratory patterns; and supervised deep learning (DL) segmentation methods require large amounts of labelled data for training. To address these issues, we propose an unsupervised breathing cycle-aware segmentation method tailored for PVA detection. Leveraging the quasi-periodic nature of ventilation waveforms, the proposed segmentation method integrates frequency-adaptive clustering, periodicity hints validation, and dynamic segmentation to identify breathing cycle boundaries. We evaluate the proposed breathing cycle-aware segmentation method on a real-world dataset from Austin Health, Melbourne, Australia, where it outperforms baseline approaches on five out of six evaluation metrics. Furthermore, classification experiments using two state-of-the-art DL-based classification models confirm that accurate segmentation of breathing cycles enhances PVA detection performance. In the future, the proposed breathing cycle-aware segmentation method could be integrated into ventilation systems to support clinical decision-making and improve patient care.

Since point clouds acquired by scanners inevitably contain noise, recovering a clean version from a noisy point cloud is essential for further 3D geometry processing applications. Several data-driven approaches have been recently introduced to overcome the drawbacks of traditional filtering algorithms, such as less robust preservation of sharp features and tedious tuning for multiple parameters. Most of these methods achieve filtering by directly regressing the position/displacement of each point, which may blur detailed features and is prone to uneven distribution. In this article, we propose a novel data-driven method that explores the implicit fields. Our assumption is that the given noisy points implicitly define a surface, and we attempt to obtain a point's movement direction and distance separately based on the predicted signed distance fields (SDFs). Taking a noisy point cloud as input, we first obtain a consistent alignment by incorporating the global points into local patches. We then feed them into an encoder-decoder structure and predict a 7D vector consisting of SDFs. Subsequently, the distance can be obtained directly from the first element in the vector, and the movement direction can be obtained by computing the gradient descent from the last six elements (i.e., six surrounding SDFs). We finally obtain the filtered results by moving each point with its predicted distance along its movement direction. Our method can produce feature-preserving results without requiring explicit normals. Experiments demonstrate that our method visually outperforms state-of-the-art methods and generally produces better quantitative results than position-based methods (both learning and non-learning).

While a popular representation of 3D data, point clouds may contain noise and need filtering before use. Existing point cloud filtering methods either cannot preserve sharp features or result in uneven point distributions in the filtered output. To address this problem, this paper introduces a point cloud filtering method that considers both point distribution and feature preservation during filtering. The key idea is to incorporate a repulsion term with a data term in energy minimization. The repulsion term is responsible for the point distribution, while the data term aims to approximate the noisy surfaces while preserving geometric features. This method is capable of handling models with fine-scale features and sharp features. Extensive experiments show that our method quickly yields good results with relatively uniform point distribution.

In recent years, deep neural networks have achieved great success in various applications, particularly in visual tasks such as image classification. However, deep neural networks cannot reach their full potential when dealing with classification problems in networks. Because the network-related dataset is usually in a structured format rather than an image format, and in some cases the data scale is small to train a deep model. Therefore, we aim at another choice, which can abandon the structure of deep neural networks, but remain the powerful representation learning ability. In this paper, we propose GTC Forest, a tree-based ensemble method for network structured data classification. GTC Forest consists of two parts: the first part Multi-Grained Traversing to do representation learning in network structured data; and the second part Cascade Forest to train on small-scale dataset, as well as reducing model complexity. Experiments are conducted on user broadband dataset, which is built to guarantee users a better Internet experience. And the results prove that our model is effective, and has higher accuracy than other machine learning methods in network structured data classification.