Search

Background In recent years, the use of deep learning has become more commonplace in the biomedical field and its development will greatly assist clinical and imaging data interpretation. Most existing machine learning methods for coronary angiography analysis are limited to a single aspect. Aims We aimed to achieve an automatic and multimodal analysis to recognise and quantify coronary angiography, integrating multiple aspects, including the identification of coronary artery segments and the recognition of lesion morphology. Methods A data set of 20,612 angiograms was retrospectively collected, among which 13,373 angiograms were labelled with coronary artery segments, and 7,239 were labelled with special lesion morphology. Trained and optimised by these labelled data, one network recognised 20 different segments of coronary arteries, while the other detected lesion morphology, including measures of lesion diameter stenosis as well as calcification, thrombosis, total occlusion, and dissection detections in an input angiogram. Results For segment prediction, the recognition accuracy was 98.4%, and the recognition sensitivity was 85.2%. For detecting lesion morphologies including stenotic lesion, total occlusion, calcification, thrombosis, and dissection, the F1 scores were 0.829, 0.810, 0.802, 0.823, and 0.854, respectively. Only two seconds were needed for the automatic recognition. Conclusions Our deep learning architecture automatically provides a coronary diagnostic map by integrating multiple aspects. This helps cardiologists to flag and diagnose lesion severity and morphology during the intervention.

Traditional enhanced external counterpulsation (EECP) used for the clinical treatment of patients with coronary heart disease only assesses diastolic/systolic blood pressure (Q = D/S > 1.2). However, improvement of the hemodynamic environment surrounding vascular endothelial cells of coronary arteries after long-term application of EECP is the basis of the treatment. Currently, the quantitative hemodynamic mechanism is not well understood. In this study, a standard 0D/3D geometric multi-scale model of the coronary artery was established to simulate the hemodynamic effects of different counterpulsation modes on the vascular endothelium. In this model, the neural regulation caused by counterpulsation was thoroughly considered. Two clinical trials were carried out to verify the numerical calculation model. The results demonstrated that the increase in counterpulsation pressure amplitude and pressurization duration increased coronary blood perfusion and wall shear stress (WSS) and reduced the oscillatory shear index (OSI) of the vascular wall. However, the impact of pressurization duration was the predominant factor. The results of the standard model and the two real individual models indicated that a long pressurization duration would cause more hemodynamic risk areas by resulting in excessive WSS, which could not be reflected by the change in the Q value. Therefore, long-term pressurization during each cardiac cycle therapy is not recommended for patients with coronary heart disease and clinical treatment should not just pay attention to the change in the Q value. Additional physiological indicators can be used to evaluate the effects of counterpulsation treatment.

Magnetic resonance imaging (MRI) has been one of the most powerful and valuable imaging methods for medical diagnosis and staging of disease. Due to the long scan time of MRI acquisition, k-space under-samplings is required during the acquisition processing. Thus, MRI reconstruction, which transfers undersampled k-space data to high-quality magnetic resonance imaging, becomes an important and meaningful task. There have been many explorations on k-space interpolation for MRI reconstruction. However, most of these methods ignore the strong correlation between target slice and its adjacent slices. Inspired by this, we propose a fully data-driven deep learning algorithm for k-space interpolation, utilizing the correlation information between the target slice and its neighboring slices. A novel network is proposed, which models the inter-dependencies between different slices. In addition, the network is easily implemented and expended. Experiments show that our methods consistently surpass existing image-domain and k-space-domain magnetic resonance imaging reconstructing methods.

Distraction osteogenesis (DO) is a mechanobiological process of producing new bone and overlying soft tissues through the gradual and controlled distraction of surgically separated bone segments. The process of bone regeneration during DO is largely affected by distraction parameters. In the present study, a distraction strategy with varying distraction rates (i.e., “rate-varying distraction”) is proposed, with the aim of shortening the distraction time and improving the efficiency of DO. We hypothesized that faster and better healing can be achieved with rate-varying distractions, as compared with constant-rate distractions. A computational model incorporating the viscoelastic behaviors of the callus tissues and the mechano-regulatory tissue differentiation laws was developed and validated to predict the bone regeneration process during DO. The effect of rate-varying distraction on the healing outcomes (bony bridging time and bone formation) was examined. Compared to the constant low-rate distraction, a low-to-high rate-varying distraction provided a favorable mechanical environment for angiogenesis and bone tissue differentiation, throughout the distraction and consolidation phase, leading to an improved healing outcome with a shortened healing time. These results suggest that a rate-varying clinical strategy could reduce the overall treatment time of DO and decrease the risk of complications related to the external fixator.

The scale of training data is significant in segmentation task, especially in segmenting the medical coronary artery angiograms. Traditional semantic segmentation networks have been restricted in this field, due to the particularity of cardiac coronary angiography data, that is, it is very difficult to balance the manual labeling costs and network accuracy. On the basis of these observations, we propose a new method to generate the so-called ‘pseudo-precise’ label and a complementary training pipeline, which can improve the performance of the networks on the premise of reducing labor costs as much as possible. Our method can thus increase the f1-score by 4%-11% with the same amount of precisely labeled data.

Function magnetic resonance imaging (fMRI) data are typically contaminated by noise introduced by head motion, physiological noise, and thermal noise. To mitigate noise artifact in fMRI data, a variety of denoising methods have been developed by removing noise factors derived from the whole time series of fMRI data and therefore are not applicable to real-time fMRI data analysis. In the present study, we develop a generally applicable, deep learning based fMRI denoising method to generate noise-free realistic individual fMRI volumes (time points). Particularly, we develop a fully data-driven 3D convolutional encapsulated Long Short-Term Memory (3DConv-LSTM) approach to generate noise-free fMRI volumes regularized by an adversarial network that makes the generated fMRI volumes more realistic by fooling a critic network. The 3DConv-LSTM model also integrates a gate-controlled self-attention model to memorize short-term dependency and historical information within a memory pool. We have evaluated our method based on both task and resting state fMRI data. Both qualitative and quantitative results have demonstrated that the proposed method outperformed state-of-the-art alternative deep learning methods.

In cardiac arterial interventional therapy, coronary angiograms provides key information to physicians for treatment strategy selection. However, it is hard to extract the information about arteries with CTO lesion in absence of contrast opacification and visualization of the CTO coronary arteries in angiograms. In this paper, we present an algorithm, which can predict the extension direction of artery with CTO and reconstruct the morphology of blocked artery in coronary angiograms. First, our algorithm segment the CTO artery angiogram and extract the artery skeleton. Second, an iterative approach is preformed to reconstruct the skeleton of blocked artery. Finally, our algorithm generates the simulated postoperative angiogram according to the skeleton image. The results demonstrate that automatic morphology reconstructing of CTO coronary artery with high accuracy and reliability is feasible.

Cardiovascular disease (CVD) is one of the diseases with the highest mortality rate in modern society, while chronic total occlusion (CTO) is the initial factor that influences the success rate of percutaneous coronary intervention (PCI), which is one of the most common treatments for CVD. In this work, novel deep convolutional neural networks (CNNs) are proposed to detect the entry point of CTO and classify its morphology according to the coronary angiography automatically. Specifically, feature pyramid networks (FPN) module and model fusion technique are applied to the detection network, and data augmentation and attentive regularization loss by reciprocative learning algorithm are used in classification network. Extracted from contrary angiography, the dataset consists of 2059 cases annotated by professional cardiologists. Experiment results show that the recall of CTO detection can reach up to 89.3%, and the sensitivity and specificity of CTO classification can reach up to 94.5% and 89.1% respectively.

In the field of cardiac arterial interventional therapy, coronary angiography imaging provides key information to physicians for treatment strategy selection, while the lesion identification process is time-consuming and error-prone even for experienced doctors. This paper proposes a method for the automatic detection of lesion in cardiac coronary angiography based on the deep learning and convolution neural network for the very first time. We used 2925 medical images for building the model. Several lesions exist on the vessel of each image. We will regard these lesion areas as objects that are different from other background areas. We designed a model based on the convolution neural network, applying some advanced building block including CReLU, Inception and other advanced technology such as batch normalization, residual connections, skip-layer connection in our network model. After training, the network model can distinguish the difference between a lesion area and a normal vessel area (background), which can detect the location of the coronary artery lesion in real time without any manual intervention. For the stenosis lesion, the recall rate of detection achieves 0.88.

Artificial intelligence (AI) has been applied successfully in various fields related to medical images. However, the use of AI-aided imaging system in percutaneous cardiovascular interventions has never been reported. We sought to utilize intelligent method to recognize coronary angiography by deep