A pilot study of a cardiovascular virtual endoscopy system based on multi‑detector computed tomography in diagnosing tetralogy of Fallot in pediatric patients

Abstract

The present study aimed to investigate the capabilities of the cardiovascular virtual endoscopy (VE) system in diagnosing tetralogy of Fallot (TOF) and performing measurements. A total of 37 patients underwent two-dimensional echocardiography (2-DE) and multi-detector computed tomography (MDCT) examinations. The obtained MDCT images were applied to a cardiovascular VE system. Diagnostic time by VE was first studied and compared with MDCT. Subsequently, with surgical findings as the ground truth, the capabilities of VE, 2-DE and MDCT in diagnosing TOF and its complications were investigated. Additionally, measurements on aorta overriding ratio and diameters for the left pulmonary artery, right pulmonary artery and right ventricular outflow tract by 2-DE and VE were analyzed. Diagnostic time by VE was significantly shorter than MDCT (188±42 vs. 303±42 sec, respectively; P<0.0001). VE, MDCT and 2-DE demonstrated comparable diagnostic rates of TOF (35/37 vs. 34/37 vs. 32/37, respectively; P>0.05). Similar findings were demonstrated in diagnosing complications of the muscular ventricular septal defects, patent ductus arteriosus, vagus subclavian artery, right arch, double superior vena cava and pulmonary artery. Furthermore, in diagnosing the atrial septal defect, 2-DE outperformed MDCT and VE (accuracy, 100 vs. 81 vs. 73%, respectively; all P<0.05). In performing relevant measurements, VE outperformed MDCT and 2-DE, particularly in accessing aorta overriding ratios with no intra-operator difference (P=0.3770) and high consistency (r=0.916). In conclusion, cardiovascular VE was demonstrated to have acceptable accuracy in diagnosing TOF, and possess advantages in shortening the diagnostic time and in performing measurements.

Detection of 3D Arterial Centerline Extraction in Spiral CT Coronary Angiography

“…Subsequently, the full convolutional segmentation network VGG-seg is proposed, which is based on a modification of the VGG-16 network by removing the final fully connected layer and deconvoluting after each pooling layer, and linearly summing the results of the deconvolution of each output layer separately to classify each pixel. In the learning process, the pretraining results on ImageNet were first used as the starting network, and then this initialized model was applied to the fully convolutional segmentation network, and 109 X-ray angiography sequences from the training set were used for training to extract blood vessels; finally, the model was tested using 40 coronary angiography sequences from the test set to achieve real-time vessel segmentation [ 30 ]. In deep learning, the number of samples is generally in great demand, and the more the number of samples and the richer the variability, the stronger the generalization ability of the model and the better the effect of the network model obtained by training.…”