3D Deep Convolutional Neural Network Research Articles

Multi-Slice Dense-Sparse Learning for Efficient Liver and Tumor Segmentation.

Accurate automatic liver and tumor segmentation plays a vital role in treatment planning and disease monitoring. Recently, deep convolutional neural network (DCNNs) has obtained tremendous success in 2D and 3D medical image segmentation. However, 2D DCNNs cannot fully leverage the inter-slice information, while 3D DCNNs are computationally expensive and memory intensive. To address these issues, we first propose a novel dense-sparse training flow from a data perspective, in which, densely adjacent slices and sparsely adjacent slices are extracted as inputs for regularizing DCNNs, thereby improving the model performance. Moreover, we design a 2.5D light-weight nnU-Net from a network perspective, in which, depthwise separable convolutions are adopted to improve the efficiency. Extensive experiments on the LiTS dataset have demonstrated the superiority of the proposed method.Clinical relevance- The proposed method can effectively segment livers and tumors from CT scans with low complexity, which can be easily implemented into clinical practice.

Can 3D artificial intelligence models outshine 2D ones in the detection of intracranial metastatic tumors on magnetic resonance images?

This study aimed to compare the prediction performance of two-dimensional (2D) and three-dimensional (3D) semantic segmentation models for intracranial metastatic tumors with a volume ≥ 0.3 mL. We used postcontrast T1 whole-brain magnetic resonance (MR), which was collected from Taipei Veterans General Hospital (TVGH). Also, the study was approved by the institutional review board (IRB) of TVGH. The 2D image segmentation model does not fully use the spatial information between neighboring slices, whereas the 3D segmentation model does. We treated the U-Net as the basic model for 2D and 3D architectures. For the prediction of intracranial metastatic tumors, the area under the curve (AUC) of the 3D model was 87.6% and that of the 2D model was 81.5%. Building a semantic segmentation model based on 3D deep convolutional neural networks might be crucial to achieve a high detection rate in clinical applications for intracranial metastatic tumors.

A deep dense residual network with reduced parameters for volumetric brain tissue segmentation from MR images

Deep convolutional neural networks (DCNN) have proven to be the state-of-the-art methods for brain tissue segmentation; however, their complex architectures, and the large number of parameters make them computationally expensive and difficult to optimize. In this paper, a novel 3D DCNN architecture, which is built upon the U-Net structure, is presented for compact feature representation and efficient parameter reduction in order to segment the brain tissues into white matter, gray matter, and cerebrospinal fluid (Code is available at:https://github.com/basnetr/U-DenseResNet). The basic idea in the proposed method is to use densely connected convolutional layers and residual skip-connections in order to increase the representation capacity, improve the gradient flow, facilitate easier and better learning, and reduce the number of parameters of the network. The loss functions, cross-entropy, dice similarity, and a combination of the two are used for the training of the proposed network. Experimental results show that the proposed approach provides the best performance on the test dataset of the single-modality IBSR18 dataset containing MR scans of diverse age groups and competitive performance on the multi-modality brain tissue segmentation challenge, iSeg-2017, containing MR scans of infants while reducing, for both the datasets, the parameters ranging from 40% to 98% compared to that of the other deep-learning based architectures. The proposed method significantly reduces the number of parameters of DCNNs while still providing high degree of accuracy. The proposed method can be used for the study of brain structure and development, in detecting a wide range of abnormal tissues, to aid diagnosis, and for guiding surgical procedures.

PO-1657 Generation of synthetic CT with 3D deep convolutional neural networks for brain MR-only radiotherapy

PO-1657 Generation of synthetic CT with 3D deep convolutional neural networks for brain MR-only radiotherapy

Segmentation of focal cortical dysplasia lesions from magnetic resonance images using 3D convolutional neural networks

Computer-aided diagnosis using advanced Artific ial Intelligence (AI) techniques has become much popular over the last few years. This work automates the segmentation of Focal Cortical Dysplasia (FCD) lesions from three-dimensional (3D) Magnetic Resonance (MR) images. FCD is a type of neuronal malformation in the brain cortex and is the leading cause of intractable epilepsy, irrespective of gender or age differences. Since the neuron related abnormalities are usually resistant to drug therapy, surgical resection has been the main treatment approach for patients with intractable epilepsy. Automating the identification and segmentation of FCD is useful for neuroradiologists in pre-surgical evaluations. Convolutional Neural Networks (CNNs) have the ability to learn appropriate features from the training data without any human intervention. But, most of the state-of-the-art FCD segmentation approaches use two-dimensional (2D) CNN models despite the availability of 3D Magnetic resonance imaging (MRI) volumes, and hence fail to leverage the inter-slice information present in the MRI volumes. The major hurdles in considering a 3D CNN model are the need for a large 3D dataset, big memory, and high computation cost. A deep 3D CNN segmentation model, which can extract inter-slice information and overcomes the drawbacks of conventional 3D CNN methods to an extent, is proposed in this paper. The model uses a 3D version of U-Net with residual blocks that works on shallow depth 3D sub-volumes generated from MRI volumes. The proposed method shows superior performance over the state-of-the-art FCD segmentation methods in both qualitative and quantitative analysis.

Neural-network model for force prediction in multi-principal-element alloys

Atomistic simulations can provide useful insights into the physical properties of multi-principal-element alloys. However, classical potentials mostly fail to capture key quantum (electronic-structure) effects. We present a deep 3D convolutional neural network (3D CNN) based framework combined with a voxelization technique to design interatomic potentials for chemically complex alloys. We highlight the performance of the 3D CNN model and its efficacy in computing potentials using the medium-entropy alloy TaNbMo. In order to provide insights into the effect of voxel resolution, we implemented two approaches based on the inner and outer bounding boxes. An efficient 3D CNN model, which is as accurate as the density-functional theory (DFT) approach, for calculating potentials will provide a promising schema for accurate atomistic simulations of structure and dynamics of general multi-principle element alloys.

Predicting Localized Primordial Star Formation with Deep Convolutional Neural Networks

Abstract We investigate applying 3D deep convolutional neural networks as fast surrogate models of the formation and feedback effects of primordial stars in hydrodynamic cosmological simulations of the first galaxies. Here, we present the surrogate model to predict localized primordial star formation; the feedback model will be presented in a subsequent paper. The star formation prediction model consists of two submodels: the first is a 3D volume classifier that predicts which (10 comoving kpc)3 volumes will host star formation, followed by a 3D Inception-based U-net voxel segmentation model that predicts which voxels will form primordial stars. We find that the combined model predicts primordial star-forming volumes with high skill, with F 1 > 0.995 and true skill score (TSS) >0.994. The star formation is localized within the volume to ≲53 voxels (∼1.6 comoving kpc3) with F 1 > 0.399 and TSS >0.857. Applied to simulations with low spatial resolution, the model predicts star-forming regions in the same locations and at similar redshifts as sites in resolved full-physics simulations that explicitly model primordial star formation and feedback. When applied to simulations with lower mass resolution, we find that the model predicts star-forming regions at later redshift due to delayed structure formation resulting from lower mass resolution. Our model predicts primordial star formation without halo finding, so it will be useful in spatially under-resolved simulations that cannot resolve primordial star-forming halos. To our knowledge, this is the first model that can predict primordial star-forming regions that match highly resolved cosmological simulations.

A Semi-Automatic Step-by-Step Expert-Guided LI-RADS Grading System Based on Gadoxetic Acid-Enhanced MRI.

PurposeLiver imaging reporting and data system (LI-RADS) classification, especially the identification of LR-3 to 5 lesions with hepatocellular carcinoma (HCC) probability, is of great significance to treatment strategy determination. We aimed to develop a semi-automatic LI-RADS grading system on multiphase gadoxetic acid-enhanced MRI using deep convolutional neural networks (CNN).Patients and MethodsAn internal data set of 439 patients and external data set of 71 patients with suspected HCC were included and underwent gadoxetic acid-enhanced MRI. The expert-guided LI-RADS grading system consisted of four deep 3D CNN models including a tumor segmentation model for automatic diameter estimation and three classification models of LI-RADS major features including arterial phase hyper-enhancement (APHE), washout and enhancing capsule. An end-to-end learning system comprising single deep CNN model that directly classified the LI-RADS grade was developed for comparison.ResultsOn internal testing set, the segmentation model reached a mean dice of 0.84, with the accuracy of mapped diameter intervals as 82.7% (95% CI: 74.4%, 91.7%). The area under the curves (AUCs) were 0.941 (95% CI: 0.914, 0.961), 0.859 (95% CI: 0.823, 0.890) and 0.712 (95% CI: 0.668, 0.754) for APHE, washout and capsule, respectively. The expert-guided system significantly outperformed the end-to-end system with a LI-RADS grading accuracy of 68.3% (95% CI: 60.8%, 76.5%) vs 55.6% (95% CI: 48.8%, 63.0%) (P<0.0001). On external testing set, the accuracy of mapped diameter intervals was 91.5% (95% CI: 81.9%, 100.0%). The AUCs were 0.792 (95% CI: 0.745, 0.833), 0.654 (95% CI: 0.602, 0.703) and 0.658 (95% CI: 0.606, 0.707) for APHE, washout and capsule, respectively. The expert-guided system achieved an overall grading accuracy of 66.2% (95% CI: 58.0%, 75.2%), significantly higher than the end-to-end system of 50.1% (95% CI: 43.1%, 58.1%) (P<0.0001).ConclusionWe developed a semi-automatic step-by-step expert-guided LI-RADS grading system (LR-3 to 5), superior to the conventional end-to-end learning system. This deep learning-based system may improve workflow efficiency for HCC diagnosis in clinical practice.

Abstract P494: Automatically Predicting Modified Treatment in Cerebral Ischemia Scores From Patient Digital Subtraction Angiography Using Deep Learning

Introduction: We propose a new method for quantifying the effect of endovascular therapy for acute ischemic stroke. Currently, an mTICI (modified treatment in cerebral ischemia) score is assigned manually to document the success of endovascular revascularization therapy. The mTICI score based on Digital Subtraction Angiography (DSA), due to visual assignment, has limitations in settings where standardization is pertinent. Methods: We hypothesize that mTICI scores can be classified successfully by deep learning and thus be used as a standardized imaging biomarker. We aim to develop a regression framework using classification models that can assign continuous score to patients depending on the success of therapy, resulting in a score that is more granular than the mTICI. We use deep learning and 3D Convolutional Neural Networks (CNN) to classify frontal post-intervention DSA 2D time series into the mTICI score categories of 0, 1, 2a, 2b, and 3. An mTICI score of 0 represents no perfusion and a score of 3 represents full perfusion. The DSA series serve as features where the time dimension is the third dimension for the CNN. For our preliminary research we have condensed our groupings into binary {0,1} (0 refers to mTICI of 0, 1, 2a while 1 refers to mTICI of 2b, 3) of frontal DSA to see if Deep Learning models can categorize between the different mTICI classes. Results: We reduced our original data size of 181 patients to 93 patients in binary group 0 and 88 patients in group 1. Using a train/test split of 0.2, we have achieved a test classification accuracy of 73%, and F1-Score of 72.2% on the binary dataset. This is a good statistical indication that neural networks are able to classify between DSA. Conclusion: Neural network models show promise as a method of distinguishing between DSA to be used as an automatic standardized scoring method for acute ischemic stroke procedures. We aim to expand this research to frontal and lateral DSA images to get more vascular information to improve model accuracies. We propose using the softmax score of the classifier as a new score which will be a standardized measurement for endovascular therapy success.

A deep learning-based lung nodule density classification and segmentation method and its effectiveness under different CT reconstruction algorithms

Objective: To evaluate the diagnostic value of the lung nodule classification and segmentation algorithm based on deep learning among different CT reconstruction algorithms. Methods: Chest CT of 363 patients from June 2019 to September 2019 in Radiology Department of Tianjin Medical University Chu Hsien-I Memorial Hospital were retrospectively collected in this study, each of which consisted of images by three different reconstruction methods (lung reconstruction, mediastinal reconstruction, bone reconstruction).These collected data were used as testing set and a total of 4 185 Chest CTs including the public data set and the constructed private data set were used as the training set. A model combines 3D deep convolutional neural network and recurrent neural network under a multi-task joint learning algorithm for lung nodule classification and segmentation were constructed. The well-trained method was tested on 363 test cases using two metrics, i.e., the accuracy of the density classification and the Dice coefficient of nodule segmentation. The performances under three reconstruction methods were statistically analyzed according to the variance analysis among three different reconstruction methods. Results: The average classification accuracies of the nodule under three reconstruction methods were 98.67%±5.70%, 98.38%±6.61% and 97.89%±7.32%. Specifically, the accuracies of the solid nodules under three reconstruction methods were 98.79%±5.58%, 98.49%±6.89% and 97.90%±7.41% and the accuracies of the sub-solid nodules were 97.57%±10.19%, 98.52%±7.77% and 98.52%±7.77%. There was no significant difference in the classification accuracy of pulmonary nodules under three different reconstruction algorithms (all P>0.05). The average Dice coefficients of nodule segmentation was 79.87%±5.78%, 79.02%±6.04% and 79.31%±5.95%. There was no significant difference in the average Dice coefficients of nodule segmentation under three different reconstruction algorithms (all P>0.05). Conclusion: Deep learning algorithm which combined with 3D convolutional neural network and recurrent neural network has demonstrated relatively stable in classification and segmentation of lung nodules under different CT reconstruction method.

2D to 3D Evolutionary Deep Convolutional Neural Networks for Medical Image Segmentation.

Developing a Deep Convolutional Neural Network (DCNN) is a challenging task that involves deep learning with significant effort required to configure the network topology. The design of a 3D DCNN not only requires a good complicated structure but also a considerable number of appropriate parameters to run effectively. Evolutionary computation is an effective approach that can find an optimum network structure and/or its parameters automatically. Note that the Neuroevolution approach is computationally costly, even for developing 2D networks. As it is expected that it will require even more massive computation to develop 3D Neuroevolutionary networks, this research topic has not been investigated until now. In this article, in addition to developing 3D networks, we investigate the possibility of using 2D images and 2D Neuroevolutionary networks to develop 3D networks for 3D volume segmentation. In doing so, we propose to first establish new evolutionary 2D deep networks for medical image segmentation and then convert the 2D networks to 3D networks in order to obtain optimal evolutionary 3D deep convolutional neural networks. The proposed approach results in a massive saving in computational and processing time to develop 3D networks, while achieved high accuracy for 3D medical image segmentation of nine various datasets.

Preliminary study of AI-assisted diagnosis using FDG-PET/CT for axillary lymph node metastasis in patients with breast cancer

BackgroundTo improve the diagnostic accuracy of axillary lymph node (LN) metastasis in breast cancer patients using 2-[18F]FDG-PET/CT, we constructed an artificial intelligence (AI)-assisted diagnosis system that uses deep-learning technologies.Materials and methodsTwo clinicians and the new AI system retrospectively analyzed and diagnosed 414 axillae of 407 patients with biopsy-proven breast cancer who had undergone 2-[18F]FDG-PET/CT before a mastectomy or breast-conserving surgery with a sentinel lymph node (LN) biopsy and/or axillary LN dissection. We designed and trained a deep 3D convolutional neural network (CNN) as the AI model. The diagnoses from the clinicians were blended with the diagnoses from the AI model to improve the diagnostic accuracy.ResultsAlthough the AI model did not outperform the clinicians, the diagnostic accuracies of the clinicians were considerably improved by collaborating with the AI model: the two clinicians' sensitivities of 59.8% and 57.4% increased to 68.6% and 64.2%, respectively, whereas the clinicians' specificities of 99.0% and 99.5% remained unchanged.ConclusionsIt is expected that AI using deep-learning technologies will be useful in diagnosing axillary LN metastasis using 2-[18F]FDG-PET/CT. Even if the diagnostic performance of AI is not better than that of clinicians, taking AI diagnoses into consideration may positively impact the overall diagnostic accuracy.

3D multi-scale deep convolutional neural networks for pulmonary nodule detection.

With the rapid development of big data and artificial intelligence technology, computer-aided pulmonary nodule detection based on deep learning has achieved some successes. However, the sizes of pulmonary nodules vary greatly, and the pulmonary nodules have visual similarity with structures such as blood vessels and shadows around pulmonary nodules, which make the quick and accurate detection of pulmonary nodules in CT image still a challenging task. In this paper, we propose two kinds of 3D multi-scale deep convolution neural networks for nodule candidate detection and false positive reduction respectively. Among them, the nodule candidate detection network consists of two parts: 1) the backbone network part Res2SENet, which is used to extract multi-scale feature information of pulmonary nodules, it is composed of the multi-scale Res2Net modules of multiple available receptive fields at a granular level and the squeeze-and-excitation units; 2) the detection part, which uses a region proposal network structure to determine region candidates, and introduces context enhancement module and spatial attention module to improve detection performance. The false positive reduction network, also composed of the multi-scale Res2Net modules and the squeeze-and-excitation units, can further classify the nodule candidates generated by the nodule candidate detection network and screen out the ground truth positive nodules. Finally, the prediction probability generated by the nodule candidate detection network is weighted average with the prediction probability generated by the false positive reduction network to obtain the final results. The experimental results on the publicly available LUNA16 dataset showed that the proposed method has a superior ability to detect pulmonary nodules in CT images.

First Law of Motion: Influencer Video Advertising on TikTok

This paper develops an algorithm to predict the effect of influencer video advertising on product sales. We propose the concept of motion-score, or m-score, a summary statistic that captures the extent to which a product is advertised in the most engaging parts of a video. We locate product placement with an object detection algorithm, and estimate pixel-level engagement as a saliency map by fine-tuning a deep 3D convolutional neural network on video-level engagement data. M-score is then defined as the pixel-level, engagement-weighted advertising intensity of a video. We construct and evaluate the algorithm with around 40,000 influencer video ads on TikTok, the largest short video platform of the world. We leverage variation in video posting time to identify the causal effect of video ads on product sales. Videos of higher m-score indeed lift more sales. This effect is sizable, robust, and more pronounced among impulsive, hedonic, or inexpensive products. We trace the mechanism to influencers’ incentives to promote themselves rather than the product. We discuss how various stakeholders in entertainment commerce can use m-score in a scalable way to optimize content, align incentives, and improve efficiency.

A 3D Deep Learning Model for Rapid Prediction of Structural Dynamics of Workpieces During Machining

Chatter stability in machining flexible parts depends directly on the structural dynamics of the workpiece. This paper proposes a novel two-stage framework that combines finite element (FE) and data-driven deep learning techniques to rapidly predict the varying dynamics of workpieces during machining. An automated framework is developed to create a large training dataset of CAD models with gradually-changing geometries. A deep 3D Convolutional Neural Network (3D-CNN) is developed to “learn” the variations in dynamic parameters as a function of geometry. The current model has been successfully implemented for prediction of natural frequencies of workpieces during turning operations. The proposed framework can be used as a computationally efficient tool in online process monitoring and automated correction applications.

Deep symmetric three-dimensional convolutional neural networks for identifying acute ischemic stroke via diffusion-weighted images.

Acute ischemic stroke (AIS) results in high morbidity, disability, and mortality. Early and automatic diagnosis of AIS can help clinicians administer the appropriate interventions. To develop a deep symmetric 3D convolutional neural network (DeepSym-3D-CNN) for automated AIS diagnosis via diffusion-weighted imaging (DWI) images. This study includes 190 study subjects (97 AIS and 93 Non-AIS) by collecting both DWI and Apparent Diffusion Coefficient (ADC) images. 3D DWI brain images are split into left and right hemispheres and input into two paths. A map with 125×253×14×12 features is extracted by each path of Inception Modules. After the features computed from two paths are subtracted through L-2 normalization, four multi-scale convolution layers produce the final predation. Three comparative models using DWI images including MedicalNet with transfer learning, Simple DeepSym-3D-CNN (each 3D Inception Module is replaced by a simple 3D-CNN layer), and L-1 DeepSym-3D-CNN (L-2 normalization is replaced by L-1 normalization) are constructed. Moreover, using ADC images and the combination of DWI and ADC images as inputs, the performance of DeepSym-3D-CNN is also investigated. Performance levels of all three models are evaluated by 5-fold cross-validation and the values of area under ROC curve (AUC) are compared by DeLong's test. DeepSym-3D-CNN achieves an accuracy of 0.850 and an AUC of 0.864. DeLong's test of AUC values demonstrates that DeepSym-3D-CNN significantly outperforms other comparative models (p < 0.05). The highlighted regions in the feature maps of DeepSym-3D-CNN spatially match with AIS lesions. Meanwhile, DeepSym-3D-CNN using DWI images presents the significant higher AUC than that either using ADC images or using DWI-ADC images based on DeLong's test (p < 0.05). DeepSym-3D-CNN is a potential method for automatically identifying AIS via DWI images and can be extended to other diseases with asymmetric lesions.

3D multi-scale deep convolutional neural networks in pulmonary nodule detection

3D multi-scale deep convolutional neural networks in pulmonary nodule detection

Application of Lithological Mapping Based on Advanced Hyperspectral Imager (AHSI) Imagery Onboard Gaofen-5 (GF-5) Satellite

The Advanced Hyperspectral Imager (AHSI), carried by the Gaofen-5 (GF-5) satellite, is the first hyperspectral sensor that simultaneously offers broad coverage and a broad spectrum. Meanwhile, deep-learning-based approaches are emerging to manage the growing volume of data produced by satellites. However, the application potential of GF-5 AHSI imagery in lithological mapping using deep-learning-based methods is currently unknown. This paper assessed GF-5 AHSI imagery for lithological mapping in comparison with Shortwave Infrared Airborne Spectrographic Imager (SASI) data. A multi-scale 3D deep convolutional neural network (M3D-DCNN), a hybrid spectral CNN (HybridSN), and a spectral–spatial unified network (SSUN) were selected to verify the applicability and stability of deep-learning-based methods through comparison with support vector machine (SVM) based on six datasets constructed by GF-5 AHSI, Sentinel-2A, and SASI imagery. The results show that all methods produce classification results with accuracy greater than 90% on all datasets, and M3D-DCNN is both more accurate and more stable. It can produce especially encouraging results by just using the short-wave infrared wavelength subset (SWIR bands) of GF-5 AHSI data. Accordingly, GF-5 AHSI imagery could provide impressive results and its SWIR bands have a high signal-to-noise ratio (SNR), which meets the requirements of large-scale and large-area lithological mapping. And M3D-DCNN method is recommended for use in lithological mapping based on GF-5 AHSI hyperspectral data.

Predicting changes in protein thermodynamic stability upon point mutation with deep 3D convolutional neural networks.

Predicting mutation-induced changes in protein thermodynamic stability (ΔΔG) is of great interest in protein engineering, variant interpretation, and protein biophysics. We introduce ThermoNet, a deep, 3D-convolutional neural network (3D-CNN) designed for structure-based prediction of ΔΔGs upon point mutation. To leverage the image-processing power inherent in CNNs, we treat protein structures as if they were multi-channel 3D images. In particular, the inputs to ThermoNet are uniformly constructed as multi-channel voxel grids based on biophysical properties derived from raw atom coordinates. We train and evaluate ThermoNet with a curated data set that accounts for protein homology and is balanced with direct and reverse mutations; this provides a framework for addressing biases that have likely influenced many previous ΔΔG prediction methods. ThermoNet demonstrates performance comparable to the best available methods on the widely used Ssym test set. In addition, ThermoNet accurately predicts the effects of both stabilizing and destabilizing mutations, while most other methods exhibit a strong bias towards predicting destabilization. We further show that homology between Ssym and widely used training sets like S2648 and VariBench has likely led to overestimated performance in previous studies. Finally, we demonstrate the practical utility of ThermoNet in predicting the ΔΔGs for two clinically relevant proteins, p53 and myoglobin, and for pathogenic and benign missense variants from ClinVar. Overall, our results suggest that 3D-CNNs can model the complex, non-linear interactions perturbed by mutations, directly from biophysical properties of atoms.

A Three-Dimensional Deep Convolutional Neural Network for Automatic Segmentation and Diameter Measurement of Type B Aortic Dissection.

ObjectiveTo provide an automatic method for segmentation and diameter measurement of type B aortic dissection (TBAD).Materials and MethodsAortic computed tomography angiographic images from 139 patients with TBAD were consecutively collected. We implemented a deep learning method based on a three-dimensional (3D) deep convolutional neural (CNN) network, which realizes automatic segmentation and measurement of the entire aorta (EA), true lumen (TL), and false lumen (FL). The accuracy, stability, and measurement time were compared between deep learning and manual methods. The intra- and inter-observer reproducibility of the manual method was also evaluated.ResultsThe mean dice coefficient scores were 0.958, 0.961, and 0.932 for EA, TL, and FL, respectively. There was a linear relationship between the reference standard and measurement by the manual and deep learning method (r = 0.964 and 0.991, respectively). The average measurement error of the deep learning method was less than that of the manual method (EA, 1.64% vs. 4.13%; TL, 2.46% vs. 11.67%; FL, 2.50% vs. 8.02%). Bland-Altman plots revealed that the deviations of the diameters between the deep learning method and the reference standard were −0.042 mm (−3.412 to 3.330 mm), −0.376 mm (−3.328 to 2.577 mm), and 0.026 mm (−3.040 to 3.092 mm) for EA, TL, and FL, respectively. For the manual method, the corresponding deviations were −0.166 mm (−1.419 to 1.086 mm), −0.050 mm (−0.970 to 1.070 mm), and −0.085 mm (−1.010 to 0.084 mm). Intra- and inter-observer differences were found in measurements with the manual method, but not with the deep learning method. The measurement time with the deep learning method was markedly shorter than with the manual method (21.7 ± 1.1 vs. 82.5 ± 16.1 minutes, p < 0.001).ConclusionThe performance of efficient segmentation and diameter measurement of TBADs based on the 3D deep CNN was both accurate and stable. This method is promising for evaluating aortic morphology automatically and alleviating the workload of radiologists in the near future.