Automatic coronary artery calcium scoring in cardiac CT angiography using paired convolutional neural networks
Graphical abstract
Introduction
Cardiovascular disease (CVD) is the global leading cause of death. The amount of coronary artery calcification (CAC) as quantified in cardiac CT – the calcium score – is a strong and independent predictor of CVD events (Yeboah et al., 2012).
In a clinical cardiac CT exam, a calcium scoring CT (CSCT) scan and a coronary CT angiography (CCTA) scan are typically both acquired. The CCTA scan is used for stenosis detection or identification of non-calcified plaque, and the CSCT scan is used to determine the calcium score (Hecht, 2015). However, it has been shown that CAC may also be quantified in CCTA. In a study by Pavitt et al. (2014), 85% of patients with a high calcium score in CSCT also had a high calcium score in CCTA (specificity 99%). Moreover, Mylonas et al. (2014) showed excellent agreement between CVD risk categories based on calcium scoring in CCTA and categories based on calcium scoring in CSCT (Cohen’s linearly weighted ). A recent survey reported typical radiation doses of 1 mSv for CAC scoring in CSCT (Messenger et al., 2015), while modern techniques allow CCTA acquisitions with 1.5 mSv radiation dose (Al-Mallah et al., 2014). Hence, performing calcium scoring in CCTA and omitting acquisition of the CSCT scan could reduce the radiation dose of a cardiac CT examination by 40–50% (Voros and Qian, 2012).
In clinical practice, CAC is standardly quantified in CSCT by manual identification of groups of connected voxels in the coronary artery that are above a 130 HU threshold and subsequent automatic 3D region growing (Agatston et al., 1990). This procedure is not applicable to CCTA, due to intravascular contrast material that typically enhances the arterial lumen well beyond 130 HU (Figs. 1(b) and (f)). Hence, higher global detection thresholds, ranging from 320 HU (Otton et al., 2012) to 600 HU (Glodny et al., 2009) have been proposed to emulate CAC scoring in CCTA. However, these fixed thresholds do not consider variations in lumen attenuation in CCTA, which might occur depending on protocols, scanners or contrast agents (Figs. 1(c) and (g)). This variation can be taken into account by using patient-specific or scan-specific attenuation thresholds, based on HU values taken from a ROI in the ascending aorta (Mylonas et al., 2014) or the proximal coronary arteries (Pavitt et al., 2014) (Figs. 1(d) and (h)).
Manual identification of CAC in cardiac CT requires substantial expert interaction, which makes it time-consuming and infeasible for large-scale or epidemiological studies. To overcome these limitations, (semi)-automatic calcium scoring methods have been proposed for CSCT (see e.g. Išgum et al. (2007); Kurkure et al. (2010); Shahzad et al. (2013); Wolterink et al. (2015a) and Ding et al. (2015)). Wolterink et al. (In press) provide a comparison of (semi-)automatic methods for calcium scoring in cardiac CT exams. Similarly, methods have been developed for automatic calcium scoring in CCTA. These methods typically require a (semi)-automatically extracted segmentation of the coronary arteries. Based on this segmentation, CAC has been identified as deviation from a trend line through the lumen intensity (Wesarg, Khan, Firle, 2006, Ahmed, de Graaf, Broersen, Kitslaar, Oost, Dijkstra, Bax, Reiber, Scholte, 2014), as voxels in the extracted arteries with intensities above a patient-specific HU threshold (Teßmann et al., 2011), or as deviations from a model of non-calcified artery segments (Eilot and Goldenberg, 2014). Mittal et al. (2010) did not use a model or threshold to identify CAC, but trained classifiers to identify CAC lesions along an extracted coronary artery centerline. Coronary artery tree extraction methods generally show good performance, but they have been reported to fail in patients with complex anatomy, in the distal segments of the coronary arteries, in scans with motion or noise artifacts and in scans with occlusions in the coronary arteries. In addition, severe CAC deposits affect the performance of artery extraction algorithms, restricting their applicability in CAC identification (Schaap et al., 2009). Manual correction of incorrectly segmented coronary arteries is often time-consuming and tedious.
We propose identification of CAC without initial coronary artery tree extraction. In contrast to previously proposed methods, our algorithm uses supervised learning to directly identify CAC in CCTA. Supervised learning using nearest-neighbor, SVM and randomized decision tree classifiers has been previously applied to CAC identification in CSCT (e.g. (Isgum, Prokop, Niemeijer, Viergever, van Ginneken, 2012, Wolterink, Leiner, Takx, Viergever, Isgum, 2015, Shahzad, van Walsum, Schaap, Rossi, Klein, Weustink, de Feyter, van Vliet, Niessen, 2013)). However, these methods cannot be applied in CCTA, as they classify potential CAC lesions, extracted using a clinical 130 HU threshold. In CCTA, it is non-trivial to distinguish between CAC and attenuated lumen, and the application of a predefined single detection threshold to extract potential CAC lesions is not feasible. Instead, the proposed method identifies CAC voxels to segment lesions.
CAC voxel identification in CCTA is a challenging and extremely unbalanced classification problem. The proposed algorithm therefore first limits the volume-of-interest (VOI) to a bounding box around the heart, extracted using our previously proposed algorithm (de Vos et al., 2016). Thereafter, voxels in this VOI are classified using convolutional neural networks (ConvNets). Recently, ConvNets have been successfully used in natural image classification, image segmentation and object detection. In addition, they have been used in several medical image analysis tasks, for example knee cartilage segmentation (Prasoon et al., 2013) lymph node detection (Roth et al., 2014), brain tissue segmentation (Stollenga et al., 2015), and pulmonary nodule classification (Ciompi et al., 2015). In the proposed algorithm, ConvNets automatically extract texture features from triplanar 2.5D or volumetric 3D input samples, which are combined with spatial features derived from a normalized coordinate system defined in the VOI. To classify voxels as CAC or non-CAC, a pair of ConvNets is used. These ConvNets are linked by training and together are called a ConvPair. The first ConvNet identifies voxels likely to be CAC. Such voxels are further classified by the second ConvNet, which distinguishes between CAC and CAC-like negatives. We propose a purely convolutional ConvNet architecture, which allows for fast evaluation times and can be directly applied to arbitrarily sized CCTA images. In addition, we present experiments showing that combinations of different architectures can achieve higher CAC identification performance than individual architectures.
We have previously proposed a method for CAC scoring in CCTA using a combination of a ConvNet and a Random Forest classifier (Wolterink et al., 2015b). This work extends our previous work in several ways. First, the classification procedure has been modified. Our previously proposed method used a ConvNet for voxel classification and a Random Forest classifier for lesion classification. The current method uses two sequential ConvNets for voxel classification. Second, in our previous work, candidate voxels for classification were selected based on the image intensity histogram. In the current work, we classify all voxels within the VOI, regardless of intensity, hence no assumptions are made about CAC HU values. Third, location features were previously extracted using a time-consuming elastic registration preprocessing step. In the current method, this registration step is omitted in favor of our very fast ConvNet-based bounding box detection technique (de Vos et al., 2016). Fourth, in our previous work we only evaluated triplanar 2.5D input with one input size. In the current work, we provide a comparison between 2.5D and volumetric 3D input, between input with different sizes, as well as experiments with ensembles combining these input representations. Fifth, the ConvNet architecture in our previous work required a time-consuming scan algorithm with many redundant operations for neighboring candidates. Here, we use a purely convolutional network for efficient voxel classification. Finally, in this work an evaluation on a substantially larger set of scans has been performed, and a thorough comparison with clinically used CSCT CAC scores, as well as interobserver variability, are provided.
Section snippets
Data
In this study, clinically obtained cardiac CT exams of 250 consecutively scanned patients were included. Each exam consists of a CSCT and a CCTA scan, made on a 256-detector row scanner (Philips Brilliance iCT, Philips Medical, Best, The Netherlands). The CSCT scans were acquired using a standard calcium scoring protocol with 120 kVp tube voltage and 55 mAs tube current, with ECG-triggering and without contrast enhancement. Reconstructed sections had 3.0 mm spacing and thickness. The CCTA scans
Method
CAC was identified by voxel classification. Besides CAC, a typical CCTA scan contains many other voxels of appearance similar to CAC. These include extracardiac lesions like bones such as ribs, calcifications in the descending aorta and calcified lymph nodes, as well as intracardiac calcifications such as those in the mitral and aortic valve (Fig. 2). In addition, coronary artery lumen is often highly attenuated, hence resembling CAC.
The proposed algorithm is illustrated in Fig. 3. First, a
Experiments and results
The set of 250 exams was divided into four sets. First, a set of 50 exams was used to train the bounding box extraction algorithm. Second, a training set of 90 exams was used to train the ConvNet pairs for CAC classification. Third, a validation set of 10 exams was used to optimize hyperparameters of the ConvNets. Finally, a test set of 100 exams was only used to evaluate the performance of the method. For the test set, annotations by both observer O1 and O2 were available.
Discussion
A method for automatic coronary artery calcium scoring in coronary CT angiography employing convolutional neural networks has been presented. In contrast to previously proposed methods for CAC scoring in CCTA, our method does not require coronary artery extraction. Instead, CAC voxels are directly identified using pairs of ConvNets.
Automatically obtained as well as reference CAC volume scores in CCTA were lower than in CSCT. This is in accordance with previous studies (van der Bijl, Joemai,
Acknowledgments
This study was financially supported by the project FSCAD, funded by the Netherlands Organisation for Health Research and Development (ZonMw) in the framework of the research programme IMDI (Innovative Medical Devices Initiative); project 104003009.
Dr. Leiner is a recipient of a ZonMw Clinical Fellowship (2011-40-00703-98-11432).
We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Tesla K40 GPU used for this research.
References (50)
- et al.
Quantification of coronary artery calcium using ultrafast computed tomography
J. Am. Coll. Cardiol.
(1990) - et al.
Automatic classification of pulmonary peri-fissural nodules in computed tomography using an ensemble of 2D views and a convolutional neural network out-of-the-box
Med. Imag. Anal.
(2015) Coronary artery calcium scanning: Past, present, and future
JACC: cardiovasc. Imag.
(2015)- et al.
Evidence for lower variability of coronary artery calcium mineral mass measurements by multi-detector computed tomography in a community-based cohort - consequences for progression studies
Eur. J. Radiol.
(2006) - et al.
What is the role of calcium scoring in the age of coronary computed tomographic angiography?
J. Nucl. Cardiol.
(2012) - et al.
A method for coronary artery calcium scoring using contrast-enhanced computed tomography
J. Cardiovasc. Comput. Tomogr.
(2012) - et al.
SCCT guidelines on the use of coronary computed tomographic angiography for patients presenting with acute chest pain to the emergency department: a report of the Society of Cardiovascular Computed Tomography Guidelines Committee
J. Cardiovasc. Comput. Tomogr.
(2014) - et al.
Standardized evaluation methodology and reference database for evaluating coronary artery centerline extraction algorithms
Med. Imag. Anal.
(2009) - et al.
Coronary calcium scoring from contrast coronary CT angiography using a semiautomated standardized method
J. Cardiovasc. Comput. Tomogr.
(2015) - et al.
Very deep convolutional networks for large-scale image recognition
International Conference on Learning Representations
(2015)
Localizing calcifications in cardiac CT data sets using a new vessel segmentation approach
J. Digit. Imaging.
Automatic detection and quantification of the Agatston coronary artery calcium score on contrast computed tomography angiography
Int. J. Cardiovasc. Imag.
Routine low-radiation-dose coronary computed tomography angiography
Eur. Heart J. Suppl.
Assessment of Agatston coronary artery calcium score using contrast-enhanced CT coronary angiography
AJR Am. J. Roentgenol.
Coronary calcium coverage score: determination, correlates, and predictive accuracy in the multi-ethnic study of atherosclerosis
Radiology
2D image classification for 3D anatomy localization; employing deep convolutional neural networks
SPIE Medical Imaging
Automated coronary artery calcium scoring from non-contrast ct using a patient-specific algorithm
SPIE Medical Imaging
Lumen diameter of normal human coronary arteries. influence of age, sex, anatomic variation, and left ventricular hypertrophy or dilation.
Circulation
Fully automatic model-based calcium segmentation and scoring in coronary CT angiography
Int. J. Comput. Assist. Radiol. Surg.
Scene parsing with multiscale feature learning, purity trees, and optimal covers
Proceedings of the International Conference on Machine Learning (ICML’12)
Automatic heart isolation for CT coronary visualization using graph-cuts
IEEE International Symposium on Biomedical Imaging (ISBI)
A method for calcium quantification by means of CT coronary angiography using 64-multidetector CT: very high correlation with Agatston and volume scores
Eur. Radiol.
Understanding the difficulty of training deep feedforward neural networks
Proceedings of the 13th International Conference on Artificial Intelligence and Statistics
Deep sparse rectifier networks
Proceedings of the 14th International Conference on Artificial Intelligence and Statistics
Cited by (239)
The Role of Artificial Intelligence in Cardiac Imaging
2024, Radiologic Clinics of North AmericaMeasurement and Application of Incidentally Detected Coronary Calcium: JACC Review Topic of the Week
2024, Journal of the American College of CardiologyFeasibility of artificial intelligence its current status, clinical applications, and future direction in cardiovascular disease
2024, Current Problems in CardiologyArtificial Intelligence in Nuclear Cardiology: An Update and Future Trends
2024, Seminars in Nuclear Medicine2022 Artificial intelligence primer for the nuclear cardiologist
2023, Journal of Nuclear CardiologyArtificial Intelligence, Computational Simulations, and Extended Reality in Cardiovascular Interventions
2023, JACC: Cardiovascular Interventions