Phone: +31 88 75 56682
Nikolas studied Biomedical Engineering at the University of Lübeck, Germany. For his Bachelor’s thesis, he worked on X-ray based tracking to improve transbronchial biopsy of pulmonary nodules. He then spent one year at Philips Research in Hamburg, where he worked on automatic analysis of chest CT images for the detection of pulmonary embolism and on automatic segmentation of the pulmonary lobes. He is currently working on automatic calcium scoring and early detection of osteoporosis in lung cancer CT screening.
Papers in international journals
|1.||B.D. de Vos, J.M. Wolterink, T. Leiner, P.A. de Jong, N. Lessmann, I. Išgum. Direct automatic coronary calcium scoring in cardiac and chest CT. IEEE Transactions on Medical Imaging (in press), 2019 ( Abstract | pdf )
Cardiovascular disease (CVD) is the global leading cause of death. A strong risk factor for CVD events is the amount of coronary artery calcium (CAC). To meet demands of the increasing interest in quantification of CAC, i.e. coronary calcium scoring, especially as an unrequested finding for screening and research, automatic methods have been proposed. Current automatic calcium scoring methods are relatively computationally expensive and only provide scores for one type of CT. To address this, we propose a computationally efficient method that employs two ConvNets: the first performs registration to align the fields of view of input CTs and the second performs direct regression of the calcium score, thereby circumventing time-consuming intermediate CAC segmentation. Optional decision feedback provides insight in the regions that contributed to the calcium score. Experiments were performed using 903 cardiac CT and 1,687 chest CT scans. The method predicted calcium scores in less than 0.3 s. Intra-class correlation coefficient between predicted and manual calcium scores was 0.98 for both cardiac and chest CT. The method showed almost perfect agreement between automatic and manual CVD risk categorization in both datasets, with a linearly weighted Cohen's kappa of 0.95 in cardiac CT and 0.93 in chest CT. Performance is similar to that of state-of-the-art methods, but the proposed method is hundreds of times faster. By providing visual feedback, insight is given in the decision process, making it readily implementable in clinical and research settings.
|2.||N. Lessmann, B. van Ginneken, P.A. de Jong, I. Išgum. Iterative fully convolutional neural networks for automatic vertebra segmentation and identification. Medical Image Analysis, 2019, vol. 53, pp. 142-155 ( Abstract | pdf )
Precise segmentation and anatomical identification of the vertebrae provides the basis for automatic analysis of the spine, such as detection of vertebral compression fractures or other abnormalities. Most dedicated spine CT and MR scans as well as scans of the chest, abdomen or neck cover only part of the spine. Segmentation and identification should therefore not rely on the visibility of certain vertebrae or a certain number of vertebrae. We propose an iterative instance segmentation approach that uses a fully convolutional neural network to segment and label vertebrae one after the other, independently of the number of visible vertebrae. This instance-by-instance segmentation is enabled by combining the network with a memory component that retains information about already segmented vertebrae. The network iteratively analyzes image patches, using information from both image and memory to search for the next vertebra. To efficiently traverse the image, we include the prior knowledge that the vertebrae are always located next to each other, which is used to follow the vertebral column. The network concurrently performs multiple tasks, which are segmentation of a vertebra, regression of its anatomical label and prediction whether the vertebra is completely visible in the image, which allows to exclude incompletely visible vertebrae from further analyses. The predicted anatomical labels of the individual vertebrae are additionally refined with a maximum likelihood approach, choosing the overall most likely labeling if all detected vertebrae are taken into account. This method was evaluated with five diverse datasets, including multiple modalities (CT and MR), various fields of view and coverages of different sections of the spine, and a particularly challenging set of low-dose chest CT scans. For vertebra segmentation, the average Dice score was 94.9 ± 2.1% with an average absolute symmetric surface distance of 0.2 ± 10.1mm. The anatomical identification had an accuracy of 93 %, corresponding to a single case with mislabeled vertebrae. Vertebrae were classified as completely or incompletely visible with an accuracy of 97 %. The proposed iterative segmentation method compares favorably with state-of-the-art methods and is fast, flexible and generalizable.
|3.||N. Lessmann, P.A. de Jong, C. Celeng, R.A.P. Takx, M.A. Viergever, B. van Ginneken, I. Išgum. Sex differences in coronary artery and thoracic aorta calcification and their association with cardiovascular mortality in heavy smokers. JACC: Cardiovascular Imaging (in press), 2019 ( Abstract | pdf )
Objectives: The aim of this study was to investigate sex differences in the prevalence, extent, and association of coronary artery calcium (CAC) and thoracic aorta calcium (TAC) scores with cardiovascular mortality in a population eligible for lung screening.
Background: CAC and TAC scores derived from chest computed tomography (CT) might be useful biomarkers for individualized cardiovascular disease prevention and could be especially relevant in high-risk populations such as heavy smokers. Therefore, it is important to know the prevalence of arterial calcifications in male and female heavy smokers, and if there are differences in the predictive value calcifications carry.
Methods: We performed a nested case–control study with 5,718 participants of the CT arm of the NLST (National Lung Screening Trial). Prevalence and extent of CAC and TAC were resampled to the full cohort to provide unbiased estimates of the typical calcium burden of male and female heavy smokers. Weighted Cox proportional hazards regression was used to assess differences in the association of CAC and TAC scores with all-cause and cardiovascular mortality.
Results: CAC was substantially more common and more severe in men (prevalence: 81% vs. 60%; median volume: 104 mm³ vs.12 mm³). Women had CAC comparable to that of men who were 10 years younger. TAC was equally common in men and women, with a tendency to be more pronounced in women (prevalence: 92% vs. 93%; median volume: 388 mm³ vs. 404 mm³). Both types of calcification were associated with increased cardiovascular and all-cause mortality. TAC scores improved the prediction of coronary heart disease mortality over CAC in men, but not in women. In both sexes, TAC, but not CAC, was associated with cardiovascular mortality other than coronary heart disease.
Conclusions: CAC develops later in women, whereas TAC develops equally in both sexes. CAC is strongly associated with coronary heart disease, whereas TAC is especially associated with extracardiac vascular mortality in either sex.
|4.||J. Šprem, B.D. de Vos, N. Lessmann, R.W. van Hamersvelt, M.J.W. Greuter, P.A. de Jong, T. Leiner, M.A. Viergever, I. Išgum. Coronary calcium scoring with partial volume correction in anthropomorphic thorax phantom and screening chest CT images. PLoS One, 2018, vol. 13, nr. 12, pp. e0209318 ( Abstract | pdf )
The amount of coronary artery calcium determined in CT scans is a well established predictor of cardiovascular events. However, high interscan variability of coronary calcium quantification may lead to incorrect cardiovascular risk assignment. Partial volume effect contributes to high interscan variability. Hence, we propose a method for coronary calcium quantification employing partial volume
Two phantoms containing artificial coronary artery calcifications and 293 subject chest CT scans were used. The first and second phantom contained nine calcifications and the second phantom contained three artificial arteries with three calcifications of different volumes, shapes and densities. The first phantom was scanned five times with and without extension rings. The second phantom was scanned three times without and with simulated cardiac motion (10 and 30 mm/s). Chest CT scans were acquired without ECG-synchronization and reconstructed using sharp and soft kernels. Coronary calcifications were annotated employing the clinically used intensity value thresholding (130 HU). Thereafter, a threshold separating each calcification from its background was determined using an Expectation-Maximization algorithm.
Finally, for each lesion the partial content of calcification in each voxel was determined depending on its intensity and the determined threshold.
Clinical calcium scoring resulted in overestimation of calcium volume for medium and high density calcifications in the first phantom, and overestimation of calcium volume for high density and underestimation for low density calcifications in the second phantom.
With induced motion these effects were further emphasized.
The proposed quantification resulted in better accuracy and substantially lower over- and underestimation of calcium volume even in presence of motion.
In chest CT, the agreement between calcium scores from the two reconstructions improved when proposed method was
Compared with clinical calcium scoring, proposed quantification provides a better estimate of the true calcium volume in phantoms and better agreement in calcium scores between different subject scan reconstructions.
|5.||J. Šprem, B.D. de Vos, N. Lessmann, P.A. de Jong, M.A. Viergever, I. Išgum. Impact of automatically detected motion artifacts on coronary calcium scoring in chest CT. Journal of Medical Imaging, 2018, vol. 5, nr. 4, pp. 044007 ( Abstract | pdf )
The amount of coronary artery calcification (CAC) quantified in CT scans enables prediction of cardiovascular disease (CVD) risk. However, interscan variability of CAC quantification is high, especially in scans made without ECG synchronization. We propose a method for automatic detection of CACs that are severely affected by cardiac motion. Subsequently, we evaluate the impact of such CACs on CAC quantification and CVD risk determination. This study includes 1000 baseline and 585 one year follow-up low-dose chest CTs from the National Lung Screening Trial. 415 baseline scans are used to train and evaluate a convolutional neural network that identifies observer determined CACs affected by severe motion artifacts. Thereafter, 585 paired scans acquired at baseline and follow-up were used to evaluate the impact of severe motion artifacts on CAC quantification and risk categorization. Based on the CAC amount the scans were categorized into four standard CVD risk categories. The method identified CACs affected by severe motion artifacts with 85.2% accuracy. Moreover, reproducibility of CAC scores in scan pairs is higher in scans containing mostly CACs not affected by severe cardiac motion. Hence, the proposed method enables identification of scans affected by severe cardiac motion where CAC quantification may not be reproducible.
|6.||S.A.M. Gernaat, S.G.M. van Velzen, V. Koh, M.J. Emaus, I. Išgum, N. Lessmann, S. Moes, A. Jacobson, P.W. Tan, D.E. Grobbee, D.H.J. van den Bongard, J.I. Tang, H.M. Verkooijen. Automatic quantification of calcifications in the coronary arteries and thoracic aorta on radiotherapy planning CT scans of Western and Asian breast cancer patients. Radiotherapy and Oncology, 2018, vol. 127, nr. 3, pp. 487-492 ( Abstract | pdf )
Purpose This study automatically quantified calcifications in coronary arteries (CAC) and thoracic aorta (TAC) on breast planning computed tomography (CT) scans and assessed its reproducibility compared to manual scoring. Material and Methods Dutch (n=1,199) and Singaporean (n=1,090) breast cancer patients with radiotherapy planning CT scan were included. CAC and TAC were automatically scored using deep learning algorithm. CVD risk categories were based on Agatson CAC: 0, 1-10, 11-100, 101-400 and >400. Reliability between automatic and manual scoring was assessed in 120 randomly selected CT scans from each population, with linearly weighted kappa for CAC categories and intraclass correlation coefficient for TAC. Results Median age was higher in Dutch patients than Singaporean patients: 57 versus 52 years. CAC and TAC increased with age and were more present in Dutch patients than Singaporean patients: 24.2% versus 17.3% and 73.0% versus 62.2%, respectively. Reliability of CAC categories and TAC was excellent in the Netherlands (0.85 (95% confidence interval (CI)=0.77-0.93) and 0.98 (95% CI=0.96-0.98) respectively) and Singapore (0.90 (95% CI=0.84-0.96) and 0.99 (95% CI=0.98-0.99) respectively). Conclusions CAC and TAC prevalence was considerable and increased with age. Deep learning software is a reliable method to automatically measure CAC and TAC on radiotherapy breast CT scans.
|7.||N. Lessmann, B. van Ginneken, M. Zreik, P.A. de Jong, B.D. de Vos, M.A. Viergever, I. Išgum. Automatic calcium scoring in low-dose chest CT using deep neural networks with dilated convolutions. IEEE Transactions on Medical Imaging, 2018, vol. 37, nr. 2, pp. 615-625 ( Abstract | pdf )
Heavy smokers undergoing screening with low-dose chest CT are affected by cardiovascular disease as much as by lung cancer. Low-dose chest CT scans acquired in screening enable quantification of atherosclerotic calcifications and thus enable identification of subjects at increased cardiovascular risk. This paper presents a method for automatic detection of coronary artery, thoracic aorta and cardiac valve calcifications in low-dose chest CT using two consecutive convolutional neural networks. The first network identifies and labels potential calcifications according to their anatomical location and the second network identifies true calcifications among the detected candidates. This method was trained and evaluated on a set of 1744 CT scans from the National Lung Screening Trial. To determine whether any reconstruction or only images reconstructed with soft tissue filters can be used for calcification detection, we evaluated the method on soft and medium/sharp filter reconstructions separately. On soft filter reconstructions, the method achieved F1 scores of 0.89, 0.89, 0.67, and 0.55 for coronary artery, thoracic aorta, aortic valve and mitral valve calcifications, respectively. On sharp filter reconstructions, the F1 scores were 0.84, 0.81, 0.64, and 0.66, respectively. Linearly weighted kappa coefficients for risk category assignment based on per subject coronary artery calcium were 0.91 and 0.90 for soft and sharp filter reconstructions, respectively. These results demonstrate that the presented method enables reliable automatic cardiovascular risk assessment in all low-dose chest CT scans acquired for lung cancer screening.
|8.||M. Zreik, N. Lessmann, R.W. van Hamersvelt, J.M. Wolterink, M. Voskuil, M.A. Viergever, T. Leiner, I. Išgum. Deep learning analysis of the myocardium in coronary CT angiography for identification of patients with functionally significant coronary artery stenosis. Medical Image Analysis, 2018, vol. 44, pp. 72-85 ( Abstract | pdf )
In patients with coronary artery stenoses of intermediate severity, the functional signicance needs to be determined. Fractional flow reserve (FFR) measurement, performed during invasive coronary angiography (ICA), is most often used in clinical practice. To reduce the number of ICA procedures, we present a method for automatic identication of patients with functionally signicant coronary artery stenoses, employing deep learning analysis of the left ventricle (LV) myocardium in rest coronary CT angiography (CCTA). The study includes consecutively acquired CCTA scans of 166 patients who underwent invasive FFR measurements. To identify patients with a functionally signicant coronary artery stenosis, analysis is performed in several stages. First, the LV myocardium is segmented using a multiscale convolutional neural network (CNN). To characterize the segmented LV myocardium, it is subsequently encoded using unsupervised convolutional autoencoder (CAE). As ischemic changes are expected to appear locally, the LV myocardium is divided into a number of spatially connected clusters, and statistics of the encodings are computed as features. Thereafter, patients are classied according to the presence of functionally signicant stenosis using an SVM classier based on the extracted features. Quantitative evaluation of LV myocardium segmentation in 20 images resulted in an average Dice coecient of 0:91 and an average mean absolute distance between the segmented and reference LV boundaries of 0:7 mm. Twenty CCTA images were used to train the LV myocardium encoder. Classication of patients was evaluated in the remaining 126 CCTA scans in 50 10-fold cross-validation experiments and resulted in an area under the receiver operating characteristic curve of 0.74 0.02. At sensitivity levels 0.60, 0.70 and 0.80, the corresponding specicity was 0.77, 0.71 and 0.59, respectively. The results demonstrate that automatic analysis of the LV myocardium in a single CCTA scan acquired at rest, without assessment of the anatomy of the coronary arteries, can be used to identify patients with functionally signicant coronary artery stenosis. This might reduce the number of patients undergoing unnecessary invasive FFR measurements.
|9.||E. Pompe, P.A. de Jong, D.A. Lynch, N. Lessmann, I. Isgum, B. van Ginneken, J.-W.J. Lammers, F.A.A. Mohamed Hoesein. Computed tomographic findings in subjects who died from respiratory disease in the National Lung Screening Trial. European Respiratory Journal, 2017, vol. 49, pp. 1601814 ( Abstract | pdf )
We evaluated the prevalence of significant lung abnormalities on computed tomography (CT) in patients who died from a respiratory illness other than lung cancer in the National Lung Screening Trial (NLST). In this retrospective case–control study, NLST participants in the CT arm who died of respiratory illness other than lung cancer were matched for age, sex, pack-years and smoking status to a surviving control. A chest radiologist and a radiology resident blinded to the outcome independently scored baseline CT scans visually and qualitatively for the presence of emphysema, airway wall thickening and fibrotic lung disease. The prevalence of CT abnormalities was compared between cases and controls by using chi-squared tests. In total, 167 participants died from a respiratory cause other than lung cancer. The prevalence of severe emphysema, airway wall thickening and fibrotic lung disease were 28.7% versus 4.8%, 26.9% versus 13.2% and 18.6% versus 0.5% in cases and controls, respectively. Radiological findings were significantly more prevalent in deaths compared with controls (all p<0.001). CT-diagnosed severe emphysema, airway wall thickening and fibrosis were much more common in NLST participants who died from respiratory disease, and CT may provide an additional means of identifying these diseases.
Papers in conference proceedings
|1.||S.G.M. van Velzen, M. Zreik, N. Lessmann, M.A. Viergever, P.A. de Jong, H.M. Verkooijen, I. Išgum. Direct prediction of cardiovascular mortality from low-dose chest CT using deep learning. In: SPIE Medical Imaging, 2019 ( Abstract | pdf )
Cardiovascular disease (CVD) is a leading cause of death in the lung cancer screening population. Chest CT scans made in lung cancer screening are suitable for identification of participants at risk of CVD. Existing methods analyzing CT images from lung cancer screening for prediction of CVD events or mortality use engineered features extracted from the images combined with patient information. In this work we propose a method that automatically predicts 5-year cardiovascular mortality directly from chest CT scans without the need for hand-crafting image features. A set of 1,583 participants of the National Lung Screening Trial was included (1,188 survivors, 395 non-survivors). Low-dose chest CT images acquired at baseline were analyzed and the follow-up time was 5 years. To limit the analysis to the heart region, the heart was first localized by our previously developed algorithm for organ localization exploiting convolutional neural networks. Thereafter, a convolutional autoencoder was used to encode the identified heart region. Finally, based on the extracted encodings subjects were classified into survivors or non-survivors using a support vector machine classifier. The performance of the method was assessed in eight cross-validation experiments with 1,433 images used for training, 50 for validation and 100 for testing. The method achieved a performance with an area under the ROC curve of 0.72. The results demonstrate that prediction of cardiovascular mortality directly from low-dose screening chest CT scans, without hand-crafted features, is feasible, allowing identification of subjects at risk of fatal CVD events.
|2.||N. Lessmann, B. van Ginneken, P.A. de Jong, I. Išgum. Iterative fully convolutional neural networks for automatic vertebra segmentation. In: Medical Imaging with Deep Learning (MIDL 2018), 2018 ( Abstract | pdf )
Precise segmentation of the vertebrae is often required for automatic detection of vertebral abnormalities. This especially enables incidental detection of abnormalities such as compression fractures in images that were acquired for other diagnostic purposes. While many CT and MR scans of the chest and abdomen cover a section of the spine, they often do not cover the entire spine. Additionally, the first and last visible vertebrae are likely only partially included in such scans. In this paper, we therefore approach vertebra segmentation as an instance segmentation problem. A fully convolutional neural network is combined with an instance memory that retains information about already segmented vertebrae. This network iteratively analyzes image patches, using the instance memory to search for and segment the first not yet segmented vertebra. At the same time, each vertebra is classified as completely or partially visible, so that partially visible vertebrae can be excluded from further analyses. We evaluated this method on spine CT scans from a vertebra segmentation challenge and on low-dose chest CT scans. The method achieved an average Dice score of 95.8% and 92.1%, respectively, and a mean absolute surface distance of 0.194 mm and 0.344 mm.
|3.||N. Lessmann, B. van Ginneken, I. Išgum. Iterative convolutional neural networks for automatic vertebra identification and segmentation in CT images. In: SPIE Medical Imaging, 2018, pp. 1057408 ( Abstract | pdf )
Segmentation and identification of the vertebrae in CT images are important initial steps for automatic analysis of the spine. This paper presents an automatic method based on iteratively applied convolutional neural networks. This approach utilizes the inherent order of the vertebral column to simplify the detection problem, so that a deep neural network can be trained with a low number of manual reference segmentations. Vertebrae are identified and segmented individually in sequential order relative to a reference vertebra. Additionally, a coarse-to-fine segmentation scheme is employed: The localization and identification of the vertebrae is first performed in low-resolution images that enable the analysis of context information. The fine segmentation is performed afterwards in the original high-resolution images. In contrast to most previous methods for vertebra segmentation, this approach is not focused on modeling shape information. The method was trained and evaluated with 15 spine CT scans from the MICCAI CSI 2014 workshop challenge. These scans cover the whole thoracic and lumbar part of the spine of healthy young adults. In contrast to a non-iterative convolutional neural network, the proposed method correctly identified all vertebrae. The method achieved a mean Dice coefficient of 0.948 and a mean surface distance of 0.29 mm and thus outperforms the best method that participated in the original challenge.
|4.||N. Lessmann, I. Isgum, A.A.A. Setio, B.D. de Vos, F. Ciompi, P.A. de Jong, M. Oudkerk, W.P.Th.M. Mali, M.A. Viergever, B. van Ginneken. Deep convolutional neural networks for automatic coronary calcium scoring in a screening study with low-dose chest CT. In: SPIE Medical Imaging, 2016, vol. 9785, pp. 978511 ( Abstract | pdf )
Coronary artery calcium (CAC) scoring can identify subjects at risk of cardiovascular events in screening programs with low-dose chest CT. We present an automatic method for CAC scoring based on deep convolutional neural networks. Candidates are extracted by intensity-based thresholding and subsequently classified by three concurrent networks that analyze three orthogonal 2D image patches per voxel. The networks consist of three convolutional steps and one fully-connected layer. In 231 subjects, this method detected on average 194.3 / 199.8mm3 CAC (sensitivity 97.2%), with 10.3mm3 false-positive volume per scan. Accuracy of cardiovascular risk category assignment was 84.4% (linearly weighted kappa 0.89).
|1.||N. Khalili, N. Lessmann, E. Turk, M.A. Viergever, M.J.N.L. Benders, I. Išgum. Brain tissue segmentation in fetal MRI using convolutional neural networks with simulated intensity inhomogeneities . International Society for Magnetic Resonance in Medicine, 27th Annual Meeting & Exhibition, 2019 ( Abstract )
Automatic brain tissue segmentation in fetal MRI is a challenging task due to artifacts such as intensity inhomogeneity, caused in particular by spontaneous fetal movements during the scan. Unlike methods that estimate the bias field to remove intensity inhomogeneity as a preprocessing step in segmentation process, we aim at improving the robustness of the segmentation method by introducing an intensity inhomogeneity augmentation (IIA). The IIA simulates various patterns of intensity inhomogeneity during the training of the segmentation network. The segmentation results demonstrate an improvement in segmentation performance when the training data is augmented with IIA.
|2.||A. Schreuder, C. Jacobs, N. Lessmann, E.T. Scholten, I. Išgum, M. Prokop, C.M. Schaefer-Prokop, B. van Ginneken. Improved lung cancer and mortality prediction accuracy using survival models based on semi-automatic CT image measurements. World Conference on Lung Cancer, 2018|
|3.||R. van Hamersvelt, M. Zreik, N. Lessmann, J. Wolterink, M. Voskuil, M.A Viergever, T. Leiner, I. Isgum. Improving Specificity of Coronary CT Angiography for the Detection of Functionally Significant Coronary Artery Disease: A Deep Learning Approach. Radiological Society of North America, 103rd Annual Meeting, 2017 ( Abstract )
PURPOSE Coronary computed tomography angiography (CCTA) is an increasingly important diagnostic tool for the detection of coronary artery disease (CAD). However, due to calcium blooming and beam hardening, specificity for diagnosing functionally significant CAD is limited. The purpose of this study was to evaluate to what extent the specificity of CCTA for detection of functionally significant CAD could be improved by combining simple stenosis grading with deep-learning based analysis of left ventricular myocardium (LVM). METHOD AND MATERIALS We retrospectively included 126 patients (77% male, 58.7±9.5 years) who underwent CCTA prior to invasive fractional flow reserve (FFR). Functionally significant CAD was defined as an invasively measured FFR value below 0.78. First, the presence and degree of coronary artery stenosis was analyzed using the CAD-RADS system. Patients without a significant stenosis reported on CCTA scans were scored as functionally non-significant. For the remaining patients, fully automatic deep learning analysis of the LVM was used to identify presence of functionally significant CAD. LVM was first segmented using a convolutional neural network and then characterized by a convolutional auto-encoder (CAE). Based on the encodings generated by the CAE a support vector machine classifier identified patients with functionally significant stenosis. Diagnostic performance of this combined analysis was evaluated and compared with patient identification based only on ≥50% stenosis degree as measured in CCTA. RESULTS FFR was significant in 64 (51%) of the patients. Sensitivity and specificity of stenosis degree reported on CCTA alone were 91% and 18%, respectively. Adding deep-learning based analysis of LVM to stenosis detection resulted in improved specificity with a slight decline in sensitivity. The combined evaluation resulted in a sensitivity of 83% and a specificity of 73%. CONCLUSION Our results show that, at the expense of only a mild sensitivity decrease, a combination of clinical stenosis evaluation and automatic LVM analysis in CCTA led to substantial increase of the specificity. CLINICAL RELEVANCE/APPLICATION Adding deep learning analysis of LVM to stenosis assessment holds the potential to substantially increase specificity of CCTA and to decrease number of patients unnecessarily referred to invasive FFR.
|4.||N. Lessmann, B. van Ginneken, P.A. de Jong, W.B. Veldhuis, M.A. Viergever, I. Isgum. Deep learning analysis for automatic calcium scoring in routine chest CT. Radiological Society of North America, 103rd Annual Meeting, 2017 ( Abstract )
Coronary artery calcium (CAC) is a robust predictor of cardiovascular events (CVE) in asymptomatic individuals. Several guidelines recommend reporting of CAC scores in ungated chest CT exams. In addition, chest CT can be used to quantify thoracic aorta calcification (TAC) and cardiac valve calcification (CVC), which may further improve prediction of CVE. This study evaluates the performance of an automatic method for scoring of CAC, TAC and CVC on routine chest CT exams.
METHOD AND MATERIALS
The study includes 290 retrospectively collected chest CTs (16/64/256 slice scanners, 0.9/1.0mm slice thickness, 0.7mm increment, 100/120kvP, 60mAs, ungated non-contrast). For calcium scoring, the scans were resampled to 3mm thick slices with 1.5mm increment. Calcifications were manually identified and labeled to define a reference standard. A deep-learning-based method employing two convolutional neural networks was used to identify and label calcifications automatically. This automatic method was trained on 1012 low-dose chest CTs from the National Lung Screening Trial. Calcifications were quantified using volume and Agatston scores. Correlation of automatic and reference scores was assessed using two-way-mixed ICC. Additionally, patients were assigned to a cardiovascular risk category based on their total CAC Agatston score (0, 1-100, 101-1000, >1000). Risk category assignment was evaluated using proportion of agreement and linearly weighted к.
16 scans (5.5%) were excluded due to insufficient image quality for manual scoring. 38.7% of the remaining patients had no CAC, 18.6% had a score 1-100, 16.4% a score 101-1000, and 26.3% a score > 1000. There was excellent correlation between automatic and manual volume scores for CAC (ICC=0.925, 95% CI: 0.904-0.941), TAC (ICC=0.991, 95% CI: 0.987-0.994), aortic valve calcifications (ICC=0.791, 95% CI: 0.734-0.836) and mitral valve calcifications (ICC=0.935, 95% CI: 0.918-0.949). Automatic and manual risk categorization agreed in 81.8% and differed by one category in 15.0% of the patients with excellent reliability (к=0.82).
Fully automatic scoring of coronary, aortic and cardiac valve calcifications highly correlates with manual scoring, even in ungated routine chest CT.
|5.||B.D. de Vos, N. Lessmann, P.A. de Jong, M.A. Viergever, I. Isgum. Direct coronary artery calcium scoring in low-dose chest CT using deep learning analysis. Radiological Society of North America, 103rd Annual Meeting, 2017 ( Abstract )
PURPOSE Coronary artery calcium (CAC) score determined in screening with low-dose chest CT is a strong and independent predictor of cardiovascular events (CVE). However, manual CAC scoring in these images is cumbersome. Existing automatic methods detect CAC lesions and thereafter quantify them. However, precise localization of lesions may not be needed to facilitate identification of subjects at risk of CVE. Hence, we have developed a deep learning system for fully automatic, real-time and direct calcium scoring circumventing the need for intermediate detection of CAC lesions. METHOD AND MATERIALS The study included a set of 1,546 baseline CT scans from the National Lung Screening Trial. Three experts defined the reference standard by manually identifying CAC lesions that were subsequently quantified using the Agatston score. The designed convolutional neural network analyzed axial slices and predicted the corresponding Agatston score. Per-subject Agatston scores were determined as the sum of per-slice scores. Each subject was assigned to one of five cardiovascular risk categories (Agatston score: 0, 1-10, 10-100, 100-400, >400). The system was trained with 75% of the scans and tested with the remaining 25%. Correlation between manual and automatic CAC scores was determined using the intra class correlation coefficient (ICC). Agreement of CVD risk categorization was evaluated using accuracy and Cohen’s linearly weighted κ. RESULTS In the 386 test subjects, the median (Q1-Q3) reference Agatston score was 54 (1-321). By the reference, 95, 37, 86, 94 and 75 subjects were assigned to 0, 1-10, 10-100, 100-400, >400 risk categories, respectively. The ICC between the automatic and reference scores was 0.95. The method assigned 85% of subjects to the correct risk category with a κ of 0.90. The score was determined in <2 seconds per CT. CONCLUSION Unlike previous automatic CAC scoring methods, the proposed method allows for quantification of coronary calcium burden without the need for intermediate identification or segmentation of separate CAC lesions. The system is robust and performs analysis in real-time. CLINICAL RELEVANCE/APPLICATION The proposed method may allow real-time identification of subjects at risk of a CVE undergoing CT-based lung cancer screening without the need for intermediate segmentation of coronary calcifications.
|6.||M. Zreik, N. Lessmann, R. van Hamersvelt, J. Wolterink, M. Voskuil, M.A Viergever, T. Leiner, I. Isgum. Deep learning analysis of the left ventricular myocardium in cardiac CT images enables detection of functionally significant coronary artery stenosis regardless of coronary anatomy. Radiological Society of North America, 103rd Annual Meeting, 2017 ( Abstract )
PURPOSE Fractional flow reserve (FFR), performed during invasive coronary angiography (ICA), is the current reference standard to determine the functional significance of a coronary stenosis. Coronary Computed Tomography Angiography (CCTA) derived virtual FFR is a promising but time and computationally expensive non-invasive alternative that can reduce the number of unnecessary ICA procedures by modeling coronary artery flow dynamics. We propose a method for fully automatic identification of patients with significant coronary artery stenosis based on deep learning analysis of only the left ventricle (LV) myocardium in CCTA. METHOD AND MATERIALS The study included resting CCTA scans (Philips Brilliance iCT, 120kVp, 210-300mAs) of 166 consecutive patients (59.2 ± 9.5 years, 128 males) who underwent invasive FFR (0.79 ± 0.10). FFR provided the reference for presence of a functionally significant stenosis (cut-off 0.78) . Automatic analysis first segmented the LV myocardium using a multiscale convolutional neural network (CNN). Next, the segmented myocardium was represented with a number of encodings generated by a convolutional auto-encoder (CAE). To detect local ischemic changes, the LV myocardium was divided into a number of spatially connected clusters. Per-cluster statistics of the encodings were subsequently used by a support vector machine classifier to identify patients with functionally significant stenosis. CCTA scans of 20 patients were used to train the CNN, and an additional 20 scans were used to train the CAE. Accuracy of patient classification was evaluated using the remaining 126 CCTA scans in 50 ten-fold cross-validation experiments. In each experiment, patients were randomly assigned to training and test sets. RESULTS Classification of patients resulted in an area under the receiver operating characteristic curve of 0.74 ± 0.02. At sensitivity levels 0.60, 0.70 and 0.80, the corresponding specificity was 0.77, 0.71 and 0.59, respectively. CONCLUSION The results demonstrate that fully automatic analysis of only the LV myocardium in resting CCTA scans, without assessment of the anatomy of the coronary arteries, can be used to identify patients with functionally significant coronary artery stenosis. CLINICAL RELEVANCE/APPLICATION Deep learning analysis of the LV myocardium could increase the specificity of the clinically used visual stenosis assessment in CCTA and reduce the number of patients undergoing unnecessary ICA.
|7.||F. Mohamed Hoesein, E. Pompe, D.A. Lynch, N. Lessmann, J.W.J. Lammers, I. Isgum, P.A. de Jong. Computed tomographic findings are associated with respiratory mortality in the National Lung Screening Trial. Radiological Society of North America, 102nd Annual Meeting, 2016 ( Abstract )
PURPOSE Almost 10% of all deaths in the computed tomography (CT) arm of the National Lung Cancer Screening Trial (NLST) were due to respiratory illnesses other than lung cancer. We evaluated the importance of lung abnormalities on screening CT for survival in NLST participants. METHOD AND MATERIALS Subjects were derived from the CT-arm of the NLST that died of a respiratory illness other than lung cancer, as defined on the death certificate, matched with an equal number of control subjects, based on age, sex, pack-years, and smoking status. A chest radiologist and senior radiology resident independently and blindly scored baseline CTs for the presence of emphysema, airway wall thickening, or fibrotic lung disease. Associations between CT abnormalities and death was evaluated with a logistic regression model. RESULTS 172 died from a respiratory cause other than lung cancer. Radiologic diseases were significantly associated with higher mortality; severe emphysema OR (95%CI) 9.7 (4.6–20.4), airway wall disease OR (95%CI) 2.3 (1.3–3.9) or fibrotic lung disease OR (95%CI) 39.1 (5.1–289.6). 81 subjects were evaluated by the EVP and confirmed the diagnosis in 55 subjects. In this group, the presence of severe emphysema was significantly associated with mortality (OR=17.2, p<0.001), as well as airway remodeling (OR=3.2
|8.||N. Lessmann, I. Isgum, S. Lam, J. Mayo, P.A. de Jong, M.A. Viergever, B. van Ginneken. Automatic coronary calcium scoring and cardiovascular risk estimation in the Pan-Canadian lung cancer screening trial. Radiological Society of North America, 101th Annual Meeting, 2015 ( Abstract )
PURPOSE Coronary artery calcium (CAC) scores determined in low-dose ungated chest CT as acquired for lung cancer screening are a strong and independent predictor of cardiovascular events (CVE). Automatic CAC scoring can complement lung cancer screening by identifying subjects at risk of a CVE. We investigated agreement and reliability of an automatic CAC scoring method previously developed for CAC scoring in the Dutch-Belgian lung cancer screening trial (NELSON) in the Pan-Canadian Early Detection of Lung Cancer Study (PanCan). METHOD AND MATERIALS Our study included 149 low-dose chest CT scans from the PanCan (16x or 64x 1.0 or 1.25 mm, 120 kvP, 40-50 mAs, no IV contrast, no ECG synchronization). Prior to scoring, the scans were reconstructed to 3.1 mm slice thickness at 1.4 mm increment. In each scan, the reference standard was set by manual annotation of CAC by one observer. Only voxels with intensities above 130 HU and lesions with a minimum volume of 1.5 mm³ were considered. Subsequently, automatic CAC scoring was performed using a supervised pattern recognition method previously developed for CAC scoring in the NELSON trial. The algorithm was trained with 100 NELSON scans. Volume and Agatston scores were computed for manual and automatic scores. Subjects were assigned to a cardiovascular risk category based on the Agatston score (0-10, 11-100, 101-400, >400). Agreement was determined as proportion of agreement in risk category assignment. Reliability was determined using linearly weighted к for risk category assignment and two-way-mixed intraclass correlation coefficient (ICC) for volume scores. RESULTS Three (2.0%) scans were excluded due to metal artifacts. In the remaining scans, the reference median CAC volume was 52.3 mm3 (P25-P75: 0-287.3 mm3). 83.6% of these scans were automatically assigned to the correct risk category. Reliability of the automatic scoring was very good for both risk category assignment (к=0.83) and volume scores (ICC=0.80). CONCLUSION Automatic coronary calcium scoring in lung cancer screening CT scans is feasible. To achieve good agreement with manual scores representative training data was not necessary.