Search

Contains fulltext : 251973.pdf (Publisher’s version ) (Closed access) Amidst the ongoing pandemic, the assessment of computed tomography (CT) images for COVID-19 presence can exceed the workload capacity of radiologists. Several studies addressed this issue by automating COVID-19 classification and grading from CT images with convolutional neural networks (CNNs). Many of these studies reported initial results of algorithms that were assembled from commonly used components. However, the choice of the components of these algorithms was often pragmatic rather than systematic and systems were not compared to each other across papers in a fair manner. We systematically investigated the effectiveness of using 3-D CNNs instead of 2-D CNNs for seven commonly used architectures, including DenseNet, Inception, and ResNet variants. For the architecture that performed best, we furthermore investigated the effect of initializing the network with pretrained weights, providing automatically computed lesion maps as additional network input, and predicting a continuous instead of a categorical output. A 3-D DenseNet-201 with these components achieved an area under the receiver operating characteristic curve of 0.930 on our test set of 105 CT scans and an AUC of 0.919 on a publicly available set of 742 CT scans, a substantial improvement in comparison with a previously published 2-D CNN. This article provides insights into the performance benefits of various components for COVID-19 classification and grading systems. We have created a challenge on grand-challenge.org to allow for a fair comparison between the results of this and future research.

Contains fulltext : 235584.pdf (Publisher’s version ) (Open Access) Lung cancer computed tomography (CT) screening trials using low-dose CT have repeatedly demonstrated a reduction in the number of lung cancer deaths in the screening group compared to a control group. With various countries currently considering the implementation of lung cancer screening, recurring discussion points are, among others, the potentially high false positive rates, cost-effectiveness, and the availability of radiologists for scan interpretation. Artificial intelligence (AI) has the potential to increase the efficiency of lung cancer screening. We discuss the performance levels of AI algorithms for various tasks related to the interpretation of lung screening CT scans, how they compare to human experts, and how AI and humans may complement each other. We discuss how AI may be used in the lung cancer CT screening workflow according to the current evidence and describe the additional research that will be required before AI can take a more prominent role in the analysis of lung screening CT scans.

BackgroundA baseline computed tomography (CT) scan for lung cancer (LC) screening may reveal information indicating that certain LC screening participants can be screened less, and instead require dedicated early cardiac and respiratory clinical input. We aimed to develop and validate competing death (CD) risk models using CT information to identify participants with a low LC risk and a high CD risk.MethodsParticipant demographics and quantitative CT measures of LC, cardiovascular disease and chronic obstructive pulmonary disease were considered for deriving a logistic regression model for predicting 5-year CD risk using a sample from the National Lung Screening Trial (n=15 000). Multicentric Italian Lung Detection data were used to perform external validation (n=2287).ResultsOur final CD model outperformed an external pre-scan model (CD Risk Assessment Tool) in both the derivation (area under the curve (AUC) 0.744 (95% CI 0.727–0.761) and 0.677 (95% CI 0.658–0.695), respectively) and validation cohorts (AUC 0.744 (95% CI 0.652–0.835) and 0.725 (95% CI 0.633–0.816), respectively). By also taking LC incidence risk into consideration, we suggested a risk threshold where a subgroup (6258/23 096 (27%)) was identified with a number needed to screen to detect one LC of 216 (versus 23 in the remainder of the cohort) and ratio of 5.41 CDs per LC case (versus 0.88). The respective values in the validation cohort subgroup (774/2287 (34%)) were 129 (versus 29) and 1.67 (versus 0.43).ConclusionsEvaluating both LC and CD risks post-scan may improve the efficiency of LC screening and facilitate the initiation of multidisciplinary trajectories among certain participants.

We propose iW-Net, a deep learning model that allows for both automatic and interactive segmentation of lung nodules in computed tomography images. iW-Net is composed of two blocks: the first one provides an automatic segmentation and the second one allows to correct it by analyzing 2 points introduced by the user in the nodule's boundary. For this purpose, a physics inspired weight map that takes the user input into account is proposed, which is used both as a feature map and in the system's loss function. Our approach is extensively evaluated on the public LIDC-IDRI dataset, where we achieve a state-of-the-art performance of 0.55 intersection over union vs the 0.59 inter-observer agreement. Also, we show that iW-Net allows to correct the segmentation of small nodules, essential for proper patient referral decision, as well as improve the segmentation of the challenging non-solid nodules and thus may be an important tool for increasing the early diagnosis of lung cancer., Comment: Pre-print submitted to IEEE Transactions on Biomedical Engineering

Objective In lung cancer screening practice low-dose computed tomography, diameter, and volumetric measurement have been used in the management of screen-detected lung nodules. The aim of this study was to compare the performance of nodule malignancy risk prediction tools using diameter or volume and between computer-aided detection (CAD) and radiologist measurements. Methods Multivariable logistic regression models were prepared by using data from two multicenter lung cancer screening trials. For model development and validation, baseline low-dose computed tomography scans from the Pan-Canadian Early Detection of Lung Cancer Study and a subset of National Lung Screening Trial (NLST) scans with lung nodules 3 mm or more in mean diameter were analyzed by using the CIRRUS Lung Screening Workstation (Radboud University Medical Center, Nijmegen, the Netherlands). In the NLST sample, nodules with cancer had been matched on the basis of size to nodules without cancer. Results Both CAD-based mean diameter and volume models showed excellent discrimination and calibration, with similar areas under the receiver operating characteristic curves of 0.947. The two CAD models had predictive performance similar to that of the radiologist-based model. In the NLST validation data, the CAD mean diameter and volume models also demonstrated excellent discrimination: areas under the curve of 0.810 and 0.821, respectively. These performance statistics are similar to those of the Pan-Canadian Early Detection of Lung Cancer Study malignancy probability model with use of these data and radiologist-measured maximum diameter. Conclusion Either CAD-based nodule diameter or volume can be used to assist in predicting a nodule's malignancy risk.

Contains fulltext : 229492.pdf (Publisher’s version ) (Open Access) OBJECTIVE: To compare nodule enhancement on subtraction CT iodine maps to that on dual-energy CT iodine maps using CT datasets acquired simultaneously. METHODS: A previously-acquired set of lung subtraction and dual-energy CT maps consisting of thirty patients with 95solid pulmonary nodules (>/=4mm diameter) was used. Nodules were annotated and segmented on CT angiography, and mean nodule enhancement in the iodine maps calculated. Three radiologists scored nodule visibility with both techniques on a 4-point scale. RESULTS: Mean nodule enhancement was higher (p

Background: Accurate estimation of the malignancy risk of pulmonary nodules at chest CT is crucial for optimizing management in lung cancer screening. Purpose: To develop and validate a deep learning (DL) algorithm for malignancy risk estimation of pulmonary nodules detected at screening CT. Materials and Methods: In this retrospective study, the DL algorithm was developed with 16 077 nodules (1249 malignant) collected between 2002 and 2004 from the National Lung Screening Trial. External validation was performed in the following three cohorts collected between 2004 and 2010 from the Danish Lung Cancer Screening Trial: a full cohort containing all 883 nodules (65 malignant) and two cancer-enriched cohorts with size matching (175 nodules, 59 malignant) and without size matching (177 nodules, 59 malignant) of benign nodules selected at random. Algorithm performance was measured by using the area under the receiver operating characteristic curve (AUC) and compared with that of the Pan-Canadian Early Detection of Lung Cancer (PanCan) model in the full cohort and a group of 11 clinicians composed of four thoracic radiologists, five radiology residents, and two pulmonologists in the cancer-enriched cohorts. Results: The DL algorithm significantly outperformed the PanCan model in the full cohort (AUC, 0.93 [95% CI: 0.89, 0.96] vs 0.90 [95% CI: 0.86, 0.93]; P = .046). The algorithm performed comparably to thoracic radiologists in cancer-enriched cohorts with both random benign nodules (AUC, 0.96 [95% CI: 0.93, 0.99] vs 0.90 [95% CI: 0.81, 0.98]; P = .11) and size-matched benign nodules (AUC, 0.86 [95% CI: 0.80, 0.91] vs 0.82 [95% CI: 0.74, 0.89]; P = .26). Conclusion: The deep learning algorithm showed excellent performance, comparable to thoracic radiologists, for malignancy risk estimation of pulmonary nodules detected at screening CT. This algorithm has the potential to provide reliable and reproducible malignancy risk scores for clinicians, which may help optimize management in lung cancer screening.

An observer study showed that two of the three top-performing algorithms from a public competition (Kaggle Data Science Bowl 2017) attained performances that were not significantly worse than that of 11 radiologists for estimating lung cancer risk on low-dose CT scans. Purpose. To determine whether deep learning algorithms developed in a public competition could identify lung cancer on low-dose CT scans with a performance similar to that of radiologists. Materials and Methods. In this retrospective study, a dataset consisting of 300 patient scans was used for model assessment; 150 patient scans were from the competition set and 150 were from an independent dataset. Both test datasets contained 50 cancer-positive scans and 100 cancer-negative scans. The reference standard was set by histopathologic examination for cancer-positive scans and imaging follow-up for at least 2 years for cancer-negative scans. The test datasets were applied to the three top-performing algorithms from the Kaggle Data Science Bowl 2017 public competition: grt123, Julian de Wit and Daniel Hammack (JWDH), and Aidence. Model outputs were compared with an observer study of 11 radiologists that assessed the same test datasets. Each scan was scored on a continuous scale by both the deep learning algorithms and the radiologists. Performance was measured using multireader, multicase receiver operating characteristic analysis. Results. The area under the receiver operating characteristic curve (AUC) was 0.877 (95% CI: 0.842, 0.910) for grt123, 0.902 (95% CI: 0.871, 0.932) for JWDH, and 0.900 (95% CI: 0.870, 0.928) for Aidence. The average AUC of the radiologists was 0.917 (95% CI: 0.889, 0.945), which was significantly higher than grt123 (P = .02); however, no significant difference was found between the radiologists and JWDH (P = .29) or Aidence (P = .26). Conclusion. Deep learning algorithms developed in a public competition for lung cancer detection in low-dose CT scans reached performance close to that of radiologists.

Background The coronavirus disease 2019 (COVID-19) pandemic has spread across the globe with alarming speed, morbidity, and mortality. Immediate triage of patients with chest infections suspected to be caused by COVID-19 using chest CT may be of assistance when results from definitive viral testing are delayed. Purpose To develop and validate an artificial intelligence (AI) system to score the likelihood and extent of pulmonary COVID-19 on chest CT scans using the COVID-19 Reporting and Data System (CO-RADS) and CT severity scoring systems. Materials and Methods The CO-RADS AI system consists of three deep-learning algorithms that automatically segment the five pulmonary lobes, assign a CO-RADS score for the suspicion of COVID-19, and assign a CT severity score for the degree of parenchymal involvement per lobe. This study retrospectively included patients who underwent a nonenhanced chest CT examination because of clinical suspicion of COVID-19 at two medical centers. The system was trained, validated, and tested with data from one of the centers. Data from the second center served as an external test set. Diagnostic performance and agreement with scores assigned by eight independent observers were measured using receiver operating characteristic analysis, linearly weighted κ values, and classification accuracy. Results A total of 105 patients (mean age, 62 years ± 16 [standard deviation]; 61 men) and 262 patients (mean age, 64 years ± 16; 154 men) were evaluated in the internal and external test sets, respectively. The system discriminated between patients with COVID-19 and those without COVID-19, with areas under the receiver operating characteristic curve of 0.95 (95% CI: 0.91, 0.98) and 0.88 (95% CI: 0.84, 0.93), for the internal and external test sets, respectively. Agreement with the eight human observers was moderate to substantial, with mean linearly weighted κ values of 0.60 ± 0.01 for CO-RADS scores and 0.54 ± 0.01 for CT severity scores. Conclusion With high diagnostic performance, the CO-RADS AI system correctly identified patients with COVID-19 using chest CT scans and assigned standardized CO-RADS and CT severity scores that demonstrated good agreement with findings from eight independent observers and generalized well to external data. © RSNA, 2020 Supplemental material is available for this article.

Purpose To microsimulate the effects of three additional annual CT screening rounds on lung cancer (LC) survival in the National Lung Screening Trial (NLST). Methods We used multiple imputation to model the effect of additional screening in the full NLST cohort on the time to LC diagnosis and on LC death in those participants who were diagnosed with LC by the end of NLST. Nodule growth models were derived from a Dutch in-vivo study. Microsimulations were repeated 500 times. The method was validated by simulating three rounds of CT screening in the original chest radiography (CXR) cohort. The times up to which the simulations remained within the 95 % confidence bands of the CT cohort’s original results were used to estimate the validity of the results in the CT cohort with three additional simulated screening rounds. Results Validation of the simulation approach on the CXR cohort resulted in a LC mortality reduction which remained well within the 95 % confidence intervals of the original CT cohort up to 6.5 years after the start of simulations. Simulating additional CT screening in the CT cohort led to LCs being diagnosed earlier than originally, resulting in a relative risk reduction in LC mortality of 11 % (95 % confidence bands, 7 %–14 %) at 6.5 years. This is equivalent to preventing 71 % (48 %–94 %) more LC deaths than the original CT cohort achieved in comparison to the original CXR cohort. Conclusion Three additional annual CT screening rounds in the NLST may have led to substantial further LC mortality reduction.

Simple Summary The risk assessment of pulmonary nodules on chest computed tomography (CT) is key to the early detection of lung cancer. Whereas a nodule’s morphological characteristics are strong predictors of malignancy for that individual nodule, it remains unclear whether the frequency and CT features of pulmonary nodules contribute to the prediction accuracy of future development of LC in another location. This was found by performing risk prediction modelling using age, sex, and CT measures from lung-cancer-free scans as predictors. We found that a greater number of part-solid and ground-glass nodules in the earliest scan was linearly associated with a higher risk of lung cancer developing in another location in the future. There were also indications that CT biomarkers of other pulmonary and heart diseases are risk factors for lung cancer. Our findings endorse the utilization of information from the entire scan to improve the accuracy of lung risk assessment. Abstract The purpose of this case–cohort study was to investigate whether the frequency and computed tomography (CT) features of pulmonary nodules posed a risk for the future development of lung cancer (LC) at a different location. Patients scanned between 2004 and 2012 at two Dutch academic hospitals were cross-linked with the Dutch Cancer Registry. All patients who were diagnosed with LC by 2014 and a random selection of LC-free patients were considered. LC patients who were determined to be LC-free at the time of the scan and all LC-free patients with an adequate scan were included. The nodule count and types (solid, part-solid, ground-glass, and perifissural) were recorded per scan. Age, sex, and other CT measures were included to control for confounding factors. The cohort included 163 LC patients and 1178 LC-free patients. Cox regression revealed that the number of ground-glass nodules and part-solid nodules present were positively correlated to future LC risk. The area under the receiver operating curve of parsimonious models with and without nodule type information were 0.827 and 0.802, respectively. The presence of subsolid nodules in a clinical setting may be a risk factor for future LC development in another pulmonary location in a dose-dependent manner. Replication of the results in screening cohorts is required for maximum utility of these findings.

ObjectivesThe 2019 Lung CT Screening Reporting & Data System version 1.1 (Lung-RADS v1.1) introduced volumetric categories for nodule management. The aims of this study were to report the distribution of Lung-RADS v1.1 volumetric categories and to analyse lung cancer (LC) outcomes within 3 years for exploring personalized algorithm for lung cancer screening (LCS).MethodsSubjects from the Multicentric Italian Lung Detection (MILD) trial were retrospectively selected by National Lung Screening Trial (NLST) criteria. Baseline characteristics included selected pre-test metrics and nodule characterization according to the volume-based categories of Lung-RADS v1.1. Nodule volume was obtained by segmentation with dedicated semi-automatic software. Primary outcome was diagnosis of LC, tested by univariate and multivariable models. Secondary outcome was stage of LC. Increased interval algorithms were simulated for testing rate of delayed diagnosis (RDD) and reduction of low-dose computed tomography (LDCT) burden.ResultsIn 1248 NLST-eligible subjects, LC frequency was 1.2% at 1 year, 1.8% at 2 years and 2.6% at 3 years. Nodule volume in Lung-RADS v1.1 was a strong predictor of LC: positive LDCT showed an odds ratio (OR) of 75.60 at 1 year (p< 0.0001), and indeterminate LDCT showed an OR of 9.16 at 2 years (p= 0.0068) and an OR of 6.35 at 3 years (p= 0.0042). In the first 2 years after negative LDCT, 100% of resected LC was stage I. The simulations of low-frequency screening showed a RDD of 13.6–21.9% and a potential reduction of LDCT burden of 25.5–41%.ConclusionsNodule volume by semi-automatic software allowed stratification of LC risk across Lung-RADS v1.1 categories. Personalized screening algorithm by increased interval seems feasible in 80% of NLST eligible.Key Points• Using semi-automatic segmentation of nodule volume, Lung-RADS v1.1 selected 10.8% of subjects with positive CT and 96.87 relative risk of lung cancer at 1 year, compared to negative CT.• Negative low-dose CT by Lung-RADS v1.1 was found in 80.6% of NLST eligible and yielded 40 times lower relative risk of lung cancer at 2 years, compared to positive low-dose CT; annual screening could be preference sensitive in this group.• Semi-automatic segmentation of nodule volume and increased screening interval by volumetric Lung-RADS v1.1 could retrospectively suggest a 25.5–41% reduction of LDCT burden, at the cost of 13.6–21.9% rate of delayed diagnosis.

An observer study with seven radiologists showed that using a dedicated CT lung screening viewer (with computer-aided detection, a segmentation algorithm, and automatic Lung Imaging Reporting and Data System [Lung-RADS] categorization) improved interobserver agreement for Lung-RADS categorization while substantially reducing the reading time. Purpose. To compare the inter- and intraobserver agreement and reading times achieved when assigning Lung Imaging Reporting and Data System (Lung-RADS) categories to baseline and follow-up lung cancer screening studies by using a dedicated CT lung screening viewer with integrated nodule detection and volumetric support with those achieved by using a standard picture archiving and communication system (PACS)-like viewer. Materials and Methods. Data were obtained from the National Lung Screening Trial (NLST). By using data recorded by NLST radiologists, scans were assigned to Lung-RADS categories. For each Lung-RADS category (1 or 2, 3, 4A, and 4B), 40 CT scans (20 baseline scans and 20 follow-up scans) were randomly selected for 160 participants (median age, 61 years; interquartile range, 58-66 years; 61 women) in total. Seven blinded observers independently read all CT scans twice in a randomized order with a 2-week washout period: once by using the standard PACS-like viewer and once by using the dedicated viewer. Observers were asked to assign a Lung-RADS category to each scan and indicate the risk-dominant nodule. Inter- and intraobserver agreement was analyzed by using Fleiss k values and Cohen weighted k values, respectively. Reading times were compared by using a Wilcoxon signed rank test. Results. The interobserver agreement was moderate for the standard viewer and substantial for the dedicated viewer, with Fleiss k values of 0.58 (95% CI: 0.55, 0.60) and 0.66 (95% CI: 0.64, 0.68), respectively. The intraobserver agreement was substantial, with a mean Cohen weighted k value of 0.67. The median reading time was significantly reduced from 160 seconds with the standard viewer to 86 seconds with the dedicated viewer (P < .001). Conclusion. Lung-RADS interobserver agreement increased from moderate to substantial when using the dedicated CT lung screening viewer. The median reading time was substantially reduced when scans were read by using the dedicated CT lung screening viewer.

ObjectivesCombined assessment of cardiovascular disease (CVD), COPD and lung cancer may improve the effectiveness of lung cancer screening in smokers. The aims were to derive and assess risk models for predicting lung cancer incidence, CVD mortality and COPD mortality by combining quantitative computed tomography (CT) measures from each disease, and to quantify the added predictive benefit of self-reported patient characteristics given the availability of a CT scan.MethodsA survey model (patient characteristics only), CT model (CT information only) and final model (all variables) were derived for each outcome using parsimonious Cox regression on a sample from the National Lung Screening Trial (n=15 000). Validation was performed using Multicentric Italian Lung Detection data (n=2287). Time-dependent measures of model discrimination and calibration are reported.ResultsAge, mean lung density, emphysema score, bronchial wall thickness and aorta calcium volume are variables that contributed to all final models. Nodule features were crucial for lung cancer incidence predictions but did not contribute to CVD and COPD mortality prediction. In the derivation cohort, the lung cancer incidence CT model had a 5-year area under the receiver operating characteristic curve of 82.5% (95% CI 80.9–84.0%), significantly inferior to that of the final model (84.0%, 82.6–85.5%). However, the addition of patient characteristics did not improve the lung cancer incidence model performance in the validation cohort (CT model 80.1%, 74.2–86.0%; final model 79.9%, 73.9–85.8%). Similarly, the final CVD mortality model outperformed the other two models in the derivation cohort (survey model 74.9%, 72.7–77.1%; CT model 76.3%, 74.1–78.5%; final model 79.1%, 77.0–81.2%), but not the validation cohort (survey model 74.8%, 62.2–87.5%; CT model 72.1%, 61.1–83.2%; final model 72.2%, 60.4–84.0%). Combining patient characteristics and CT measures provided the largest increase in accuracy for the COPD mortality final model (92.3%, 90.1–94.5%) compared to either other model individually (survey model 87.5%, 84.3–90.6%; CT model 87.9%, 84.8–91.0%), but no external validation was performed due to a very low event frequency.ConclusionsCT measures of CVD and COPD provides small but reproducible improvements to nodule-based lung cancer risk prediction accuracy from 3 years onwards. Self-reported patient characteristics may not be of added predictive value when CT information is available.

Contains fulltext : 229529.pdf (Publisher’s version ) (Open Access) Several studies investigated the appearance of intrapulmonary lymph nodes (IPLNs) at CT with pathologic correlation. IPLNs are benign lesions and do not require follow-up after initial detection. There are indications that IPLNs represent a considerable portion of incidentally found pulmonary nodules seen at high-resolution CT. The reliable and accurate identification of IPLNs as benign nodules may substantially reduce the number of unnecessary follow-up CT examinations. Typical CT features of IPLNs are a noncalcified solid nodule with sharp margins; a round, oval, or polygonal shape; distanced 15 mm or less from the pleura; and most being located below the level of the carina. The term perifissural nodule (PFN) was coined based on some of these characteristics. Standardization of those CT criteria are a prerequisite for accurate nodule classification. However, four different definitions of PFNs can currently be found in the literature. Furthermore, there is considerable variation in the reported interobserver agreement, malignancy rate, and prevalence of PFNs. The purpose of this review was to provide an overview of what is known about PFNs. In addition, knowledge gaps in defining PFNs will be discussed. A decision tree to guide clinicians in classifying nodules as PFNs is provided.Supplemental material is available for this article.? RSNA, 2020See also the commentary by White and Rubin in this issue.

Purpose One of the main pathophysiological mechanisms of chronic obstructive pulmonary disease is inflammation, which has been associated with lymphadenopathy. Intrapulmonary lymph nodes can be identified on CT as perifissural nodules (PFN). We investigated the association between the number and size of PFNs and measures of COPD severity. Materials and Methods CT images were obtained from COPDGene. 50 subjects were randomly selected per GOLD stage (0 to 4), GOLD-unclassified, and never-smoker groups and allocated to either “Healthy,” “Mild,” or “Moderate/severe” groups. 26/350 (7.4%) subjects had missing images and were excluded. Supported by computer-aided detection, a trained researcher prelocated non-calcified opacities larger than 3 mm in diameter. Included lung opacities were classified independently by two radiologists as either “PFN,” “not a PFN,” “calcified,” or “not a nodule”; disagreements were arbitrated by a third radiologist. Ordinal logistic regression was performed as the main statistical test. Results A total of 592 opacities were included in the observer study. A total of 163/592 classifications (27.5%) required arbitration. A total of 17/592 opacities (2.9%) were excluded from the analysis because they were not considered nodular, were calcified, or all three radiologists disagreed. A total of 366/575 accepted nodules (63.7%) were considered PFNs. A maximum of 10 PFNs were found in one image; 154/324 (47.5%) contained no PFNs. The number of PFNs per subject did not differ between COPD severity groups (p = 0.50). PFN short-axis diameter could significantly distinguish between the Mild and Moderate/severe groups, but not between the Healthy and Mild groups (p = 0.021). Conclusions There is no relationship between PFN count and COPD severity. There may be a weak trend of larger intrapulmonary lymph nodes among patients with more advanced stages of COPD.

Pulmonary lobe segmentation in computed tomography scans is essential for regional assessment of pulmonary diseases. Recent works based on convolution neural networks have achieved good performance for this task. However, they are still limited in capturing structured relationships due to the nature of convolution. The shape of the pulmonary lobes affect each other and their borders relate to the appearance of other structures, such as vessels, airways, and the pleural wall. We argue that such structural relationships play a critical role in the accurate delineation of pulmonary lobes when the lungs are affected by diseases such as COVID-19 or COPD. In this paper, we propose a relational approach (RTSU-Net) that leverages structured relationships by introducing a novel non-local neural network module. The proposed module learns both visual and geometric relationships among all convolution features to produce self-attention weights. With a limited amount of training data available from COVID-19 subjects, we initially train and validate RTSU-Net on a cohort of 5000 subjects from the COPDGene study (4000 for training and 1000 for evaluation). Using models pre-trained on COPDGene, we apply transfer learning to retrain and evaluate RTSU-Net on 470 COVID-19 suspects (370 for retraining and 100 for evaluation). Experimental results show that RTSU-Net outperforms three baselines and performs robustly on cases with severe lung infection due to COVID-19.

Purpose: To study interreader variability for classifying pulmonary opacities at CT as perifissural nodules (PFNs) and determine how reliably radiologists differentiate PFNs from malignancies. Materials and Methods: CT studies were obtained retrospectively from the National Lung Screening Trial (2002-2009). Nodules were eligible for the study if they were noncalcified, solid, within the size range of 5 to 10 mm, and scanned with a section thickness of 2 mm or less. Six radiologists classified 359 nodules in a cancer-enriched data set as PFN, non-PFN, or not applicable. Nodules classified as not applicable by at least three radiologists were excluded, leaving 316 nodules for post-hoc statistical analysis. Results: The study group contained 22.2% cancers (70 of 316). The median proportion of nodules classified as PFNs was 45.6% (144 of 316). All six radiologists uniformly classified 17.7% (56 of 316) of the nodules as PFNs. The Fleiss k was 0.50. Compared with non-PFNs, nodules classified as PFNs were smaller and more often located in the lower lobes and attached to a fissure (P , .001). Thirteen (18.6%) of 70 cancers were misclassified 21 times as PFNs. Individual readers' misclassification rates ranged from 0% (0 of 125) to 4.9% (eight of 163). Of 13 misclassified malignancies, 11 were in the upper lobes and two were attached to a fissure. Conclusion: There was moderate interreader agreement when classifying nodules as perifissural nodules. Less than 2.5% of perifissural nodule classifications were misclassified lung cancers (21 of 865) in this cancer-enriched study. Allowing nodules classified as perifissural nodules to be omitted from additional follow-up in a screening setting could substantially reduce the number of unnecessary scans; excluding perifissural nodules in the upper lobes would greatly decrease the misclassification rate.

Machine learning (automated processes that learn by example in order to classify, predict, discover or generate new data) and artificial intelligence (methods by which a computer makes decisions or discoveries that would usually require human intelligence) are now firmly established in astronomy. Every week, new applications of machine learning and artificial intelligence are added to a growing corpus of work. Random forests, support vector machines, and neural networks (artificial, deep, and convolutional) are now having a genuine impact for applications as diverse as discovering extrasolar planets, transient objects, quasars, and gravitationally-lensed systems, forecasting solar activity, and distinguishing between signals and instrumental effects in gravitational wave astronomy. This review surveys contemporary, published literature on machine learning and artificial intelligence in astronomy and astrophysics. Applications span seven main categories of activity: classification, regression, clustering, forecasting, generation, discovery, and the development of new scientific insight. These categories form the basis of a hierarchy of maturity, as the use of machine learning and artificial intelligence emerges, progresses or becomes established., 40 pages, accepted for publication in WIREs Data Mining and Knowledge Discovery (9 November 2019)

When children learn to count, do they understand its logic independent of the number list that they learned to count with? Here we tested CP-knowers' (ages three to five) understanding of how counting reveals a set's cardinality, even when non-numerical lists are used to count. Participants watched an agent count unobservable objects in two boxes and were asked to identify the larger set. Participants successfully identified the box with more objects when the agent counted using their familiar number list (Experiment 1) and when the agent counted using a non-numeric ordered list, as long as the items in the list were not linguistically used as number words (Experiments 2–3). Additionally, children's performance was strongly influenced by visual cues that helped them link the list's order to representations of magnitude (Experiment 4). Our findings suggest that three- to six-year-olds who can count also understand how counting reveals a set's cardinality, but they require additional time to understand how symbols on any arbitrary ordered list can be used as numerals.

Contains fulltext : 181941.pdf (Publisher’s version ) (Open Access) OBJECTIVES: To compare the PanCan model, Lung-RADS and the 1.2016 National Comprehensive Cancer Network (NCCN) guidelines for discriminating malignant from benign pulmonary nodules on baseline screening CT scans and the impact diameter measurement methods have on performances. METHODS: From the Danish Lung Cancer Screening Trial database, 64 CTs with malignant nodules and 549 baseline CTs with benign nodules were included. Performance of the systems was evaluated applying the system's original diameter definitions: D(longest-C) (PanCan), D(meanAxial) (NCCN), both obtained from axial sections, and D(mean3D) (Lung-RADS). Subsequently all diameter definitions were applied uniformly to all systems. Areas under the ROC curves (AUC) were used to evaluate risk discrimination. RESULTS: PanCan performed superiorly to Lung-RADS and NCCN (AUC 0.874 vs. 0.813, p = 0.003; 0.874 vs. 0.836, p = 0.010), using the original diameter specifications. When uniformly applying D(longest-C), D(mean3D) and D(meanAxial), PanCan remained superior to Lung-RADS (p < 0.001 - p = 0.001) and NCCN (p < 0.001 - p = 0.016). Diameter definition significantly influenced NCCN's performance with D(longest-C) being the worst (D(longest-C) vs. D(mean3D), p = 0.005; D(longest-C) vs. D(meanAxial), p = 0.016). CONCLUSIONS: Without follow-up information, the PanCan model performs significantly superiorly to Lung-RADS and the 1.2016 NCCN guidelines for discriminating benign from malignant nodules. The NCCN guidelines are most sensitive to nodule size definition. KEY POINTS: * PanCan model outperforms Lung-RADS and 1.2016 NCCN guidelines in identifying malignant pulmonary nodules. * Nodule size definition had no significant impact on Lung-RADS and PanCan model. * 1.2016 NCCN guidelines were significantly superior when using mean diameter to longest diameter. * Longest diameter achieved lowest performance for all models. * Mean diameter performed equivalently when derived from axial sections and from volumetry.

Lung-RADS represents a categorical system published by the American College of Radiology to standardise management in lung cancer screening. The purpose of the study was to quantify how well readers agree in assigning Lung-RADS categories to screening CTs; secondary goals were to assess causes of disagreement and evaluate its impact on patient management. For the observer study, 80 baseline and 80 follow-up scans were randomly selected from the NLST trial covering all Lung-RADS categories in an equal distribution. Agreement of seven observers was analysed using Cohen’s kappa statistics. Discrepancies were correlated with patient management, test performance and diagnosis of malignancy within the scan year. Pairwise interobserver agreement was substantial (mean kappa 0.67, 95% CI 0.58–0.77). Lung-RADS category disagreement was seen in approximately one-third (29%, 971) of 3360 reading pairs, resulting in different patient management in 8% (278/3360). Out of the 91 reading pairs that referred to scans with a tumour diagnosis within 1 year, discrepancies in only two would have resulted in a substantial management change. Assignment of lung cancer screening CT scans to Lung-RADS categories achieves substantial interobserver agreement. Impact of disagreement on categorisation of malignant nodules was low. • Lung-RADS categorisation of low-dose lung screening CTs achieved substantial interobserver agreement. • Major cause for disagreement was assigning a different nodule as risk-dominant. • Disagreement led to a different follow-up time in 8% of reading pairs.

PurposeNormalized emphysema score is a protocol-robust CT biomarker of mortality. We aimed to improve mortality prediction by including the emphysema score progression rate-its change over time-into the models.Method and materialsCT scans from 6000 National Lung Screening Trial CT arm participants were included. Of these, 1810 died (445 lung cancer-specific). The remaining 4190 survivors were sampled with replacement up to 24432 to approximate the full cohort. Three overlapping subcohorts were formed which required participants to have images from specific screening rounds. Emphysema scores were obtained after resampling, normalization, and bullae cluster analysis of the original images. Base models contained solely the latest emphysema score. Progression models included emphysema score progression rate. Models were adjusted by including baseline age, sex, BMI, smoking status, smoking intensity, smoking duration, and previous COPD diagnosis. Cox proportional hazard models predicting all-cause and lung cancer mortality were compared by calculating the area under the curve per year follow-up.ResultsIn the subcohort of participants with baseline and first annual follow-up scans, the analysis was performed on 4940 participants (23227 after resampling). Area under the curve for all-cause mortality predictions of the base and progression models 6 years after baseline were 0.564 (0.564 to 0.565) and 0.569 (0.568 to 0.569) when unadjusted, and 0.704 (0.703 to 0.704) to 0.705 (0.704 to 0.705) when adjusted. The respective performances predicting lung cancer mortality were 0.638 (0.637 to 0.639) and 0.643 (0.642 to 0.644) when unadjusted, and 0.724 (0.723 to 0.725) and 0.725 (0.725 to 0.726) when adjusted.ConclusionIncluding emphysema score progression rate into risk models shows no clinically relevant improvement in mortality risk prediction. This is because scan normalization does not adjust for an overall change in lung density. Adjusting for changes in smoking behavior is likely required to make this a clinically useful measure of emphysema progression.

Objectives To benchmark the performance of state-of-the-art computer-aided detection (CAD) of pulmonary nodules using the largest publicly available annotated CT database (LIDC/IDRI), and to show that CAD finds lesions not identified by the LIDC's four-fold double reading process. Methods The LIDC/IDRI database contains 888 thoracic CT scans with a section thickness of 2.5 mm or lower. We report performance of two commercial and one academic CAD system. The influence of presence of contrast, section thickness, and reconstruction kernel on CAD performance was assessed. Four radiologists independently analyzed the false positive CAD marks of the best CAD system. Results The updated commercial CAD system showed the best performance with a sensitivity of 82 % at an average of 3.1 false positive detections per scan. Forty-five false positive CAD marks were scored as nodules by all four radiologists in our study. Conclusions On the largest publicly available reference database for lung nodule detection in chest CT, the updated commercial CAD system locates the vast majority of pulmonary nodules at a low false positive rate. Potential for CAD is substantiated by the fact that it identifies pulmonary nodules that were not marked during the extensive four-fold LIDC annotation process. Key Points BL CAD systems should be validated on public, heterogeneous databases. BL The LIDC/IDRI database is an excellent database for benchmarking nodule CAD. BL CAD can identify the majority of pulmonary nodules at a low false positive rate. BL CAD can identify nodules missed by an extensive two-stage annotation process.

PURPOSE: To develop a fully automated algorithm for spleen segmentation and to assess the performance of this algorithm in a large dataset. MATERIALS AND METHODS: In this retrospective study, a three-dimensional deep learning network was developed to segment the spleen on thorax-abdomen CT scans. Scans were extracted from patients undergoing oncologic treatment from 2014 to 2017. A total of 1100 scans from 1100 patients were used in this study, and 400 were selected for development of the algorithm. For testing, a dataset of 50 scans was annotated to assess the segmentation accuracy and was compared against the splenic index equation. In a qualitative observer experiment, an enriched set of 100 scan-pairs was used to evaluate whether the algorithm could aid a radiologist in assessing splenic volume change. The reference standard was set by the consensus of two other independent radiologists. A Mann-Whitney U test was conducted to test whether there was a performance difference between the algorithm and the independent observer. RESULTS: The algorithm and the independent observer obtained comparable Dice scores (P = .834) on the test set of 50 scans of 0.962 and 0.964, respectively. The radiologist had an agreement with the reference standard in 81% (81 of 100) of the cases after a visual classification of volume change, which increased to 92% (92 of 100) when aided by the algorithm. CONCLUSION: A segmentation method based on deep learning can accurately segment the spleen on CT scans and may help radiologists to detect abnormal splenic volumes and splenic volume changes. Supplemental material is available for this article. © RSNA, 2020

ObjectiveTo assess the performance of the Brock malignancy risk model for pulmonary nodules detected in routine clinical setting.MethodsIn two academic centres in the Netherlands, we established a list of patients aged ≥40 years who received a chest CT scan between 2004 and 2012, resulting in 16 850 and 23 454 eligible subjects. Subsequent diagnosis of lung cancer until the end of 2014 was established through linking with the National Cancer Registry. A nested case–control study was performed (ratio 1:3). Two observers used semiautomated software to annotate the nodules. The Brock model was separately validated on each data set using ROC analysis and compared with a solely size-based model.ResultsAfter the annotation process the final analysis included 177 malignant and 695 benign nodules for centre A, and 264 malignant and 710 benign nodules for centre B. The full Brock model resulted in areas under the curve (AUCs) of 0.90 and 0.91, while the size-only model yielded significantly lower AUCs of 0.88 and 0.87, respectively (p99%.DiscussionThe Brock model shows high predictive discrimination of potentially malignant and benign nodules when validated in an unselected, heterogeneous clinical population. The high NPV may be used to decrease the number of nodule follow-up examinations.

Contains fulltext : 193523.pdf (Publisher’s version ) (Open Access) OBJECTIVES: The aim of this study was to compare computer-aided diagnosis (CAD) and visual reading for the detection of subsolid nodules (SSNs) in volumetrl measuremic low-dose computed tomography (LDCT) for lung cancer screening. MATERIALS AND METHODS: Prospective visual detection (VD) and manuaent of SSN were performed in the 2303 baseline volumetric LDCTs of the Multicenter Italian Lung Detection trial. Baseline and 2- and 4-year LDCTs underwent retrospective CAD analysis, subsequently reviewed by 2 experienced thoracic radiologists. The reference standard was defined by the cumulative number of SSNs detected by any reading method between VD and CAD. The number of false-positive CAD marks per scan (FP/scan) was calculated. The positive predictive value of CAD was quantified per nodule (PPV) and per screenee (PPV). The sensitivity and negative predictive value were compared between CAD and VD. The longitudinal 3-time-point sensitivity of CAD was calculated in the subgroup of persistent SSNs seen by VD (ratio between the prevalent SSNs detected by CAD through 3 time points and the total number of persistent prevalent SSNs detected by VD) to test the sensitivity of iterated CAD analysis during a screening program. Semiautomatic characteristics (diameter, volume, and mass; both for whole nodule and solid component) were compared between SSN detected CAD-only or VD-only to investigate whether either reading method could suffer from specific sensitivity weakness related to SSN features. Semiautomatic and manual diameters were compared using Spearman rho correlation and Bland-Altman plot. RESULTS: Computer-aided diagnosis and VD detected a total of 194 SSNs in 6.7% (155/2,303) of screenees at baseline LDCT. The CAD showed mean FP/scan of 0.26 (604/2,303); PPV 22.5% (175/779) for any SSN, with 54.4% (37/68) for PSN and 19.4% for NSN (138/711; P < 0.001); PPV 25.6% (137/536). The sensitivity of CAD was superior to that of VD (88.4% and 34.2%, P < 0.001), as well as negative predictive value (99.2% and 95.5%, P < 0.001). The longitudinal 3-time-point sensitivity of CAD was 87.5% (42/48). There was no influence of semiautomatic characteristics on the performance of either reading method. The diameter of the solid component in PSN was larger by CAD compared with manual measurement. At baseline, CAD detected 3 of 4 SSNs, which were first overlooked by VD and subsequently evolved to lung cancer. CONCLUSIONS: Computer-aided diagnosis and VD as concurrent reading methods showed complementary performance, with CAD having a higher sensitivity, especially for PSN, but requiring visual confirmation to reduce false-positive calls. Computer-aided diagnosis and VD should be jointly used for LDCT reading to reduce false-negatives of either lone method. The semiautomatic measurement of solid core showed systematic shift toward a larger diameter, potentially resulting in an up-shift within Lung CT Screening Reporting and Data System classification.

Purpose To evaluate whether, and to which extent, experienced radiologists are able to visually correctly differentiate transient from persistent subsolid nodules from a single CT examination alone and to determine CT morphological features to make this differentiation. Materials and methods We selected 86 transient and 135 persistent subsolid nodules from the National Lung Screening Trial (NLST) database. Four experienced radiologists visually assessed a predefined list of morphological features and gave a final judgment on a continuous scale (0-100). To assess observer performance, area under the receiver operating characteristic (ROC) curve was calculated. Statistical differences of morphological features between transient and persistent lesions were calculated using Chi-square. Inter-observer agreement of morphological features was evaluated by percentage agreement. Results Forty-nine lesions were excluded by at least 2 observers, leaving 172 lesions for analysis. On average observers were able to differentiate transient from persistent subsolid nodules ≥ 10 mm with an area under the curve of 0.75 (95% CI 0.67-0.82). Nodule type, lesion margin, presence of a well-defined border, and pleural retraction showed significant differences between transient and persistent lesions in two observers. Average pair-wise percentage agreement for these features was 81%, 64%, 47% and 89% respectively. Agreement for other morphological features varied from 53% to 95%. Conclusion The visual capacity of experienced radiologists to differentiate persistent and transient subsolid nodules is moderate in subsolid nodules larger than 10 mm. Performance of the visual assessment of CT morphology alone is not sufficient to generally abandon a short-term follow-up for subsolid nodules.

Subsolid pulmonary nodules are commonly encountered in lung cancer screening and clinical routine. Compared to other nodule types, subsolid nodules are associated with a higher malignancy probability for which the size and mass of the nodule and solid core are important indicators. However, reliably measuring these characteristics on computed tomography (CT) can be hampered by the presence of vessels encompassed by the nodule, since vessels have similar CT attenuation as solid cores. This can affect treatment decisions and patient management. We present a method based on voxel classification to automatically identify vessels and solid cores in given subsolid nodules on CT. Three experts validated our method on 170 screen-detected subsolid nodules from the Multicentric Italian Lung Disease trial. The agreement between the proposed method and the observers was substantial for vessel detection and moderate for solid core detection, which was similar to the inter-observer agreement. We found a relatively high variability in the inter-observer agreement and low method-observer agreements for delineating the borders of vessels and solid cores, illustrating the difficulty of this task. However, 92.4% of the proposed vessel and 80.6% of the proposed solid core segmentations were labeled as usable in clinical practice by the majority of experts.

Contains fulltext : 191302.pdf (Publisher’s version ) (Open Access) Automatic localization of organs and other structures in medical images is an important preprocessing step that can improve and speed up other algorithms such as organ segmentation, lesion detection, and registration. This work presents an efficient method for simultaneous localization of multiple structures in 3D thorax-abdomen CT scans. Our approach predicts the location of multiple structures using a single multi-label convolutional neural network for each orthogonal view. Each network takes extra slices around the current slice as input to provide extra context. A sigmoid layer is used to perform multi-label classification. The output of the three networks is subsequently combined to compute a 3D bounding box for each structure. We used our approach to locate 11 structures of interest. The neural network was trained and evaluated on a large set of 1884 thorax-abdomen CT scans from patients undergoing oncological workup. Reference bounding boxes were annotated by human observers. The performance of our method was evaluated by computing the wall distance to the reference bounding boxes. The bounding boxes annotated by the first human observer were used as the reference standard for the test set. Using the best configuration, we obtained an average wall distance of 3.20+/-7.33mm in the test set. The second human observer achieved 1.23+/-3.39mm. For all structures, the results were better than those reported in previously published studies. In conclusion, we proposed an efficient method for the accurate localization of multiple organs. Our method uses multiple slices as input to provide more context around the slice under analysis, and we have shown that this improves performance. This method can easily be adapted to handle more organs.

BackgroundAll lung cancer CT screening trials used fixed follow-up intervals, which may not be optimal. We developed new lung cancer risk models for personalising screening intervals to 1 year or 2 years, and compared these with existing models.MethodsWe included participants in the CT arm of the National Lung Screening Trial (2002–2010) who underwent a baseline scan and a first annual follow-up scan and were not diagnosed with lung cancer in the first year. True and false positives and the area under the curve of each model were calculated. Internal validation was performed using bootstrapping.ResultsData from 24 542 participants were included in the analysis. The accuracy was 0.785, 0.693, 0.697, 0.666 and 0.727 for the polynomial, patient characteristics, diameter, Patz and PanCan models, respectively. Of the 24 542 participants included, 174 (0.71%) were diagnosed with lung cancer between the first and the second annual follow-ups. Using the polynomial model, 2558 (10.4%, 95% CI 10.0% to 10.8%), 7544 (30.7%, 30.2% to 31.3%), 10 947 (44.6%, 44.0% to 45.2%), 16 710 (68.1%, 67.5% to 68.7%) and 20 023 (81.6%, 81.1% to 92.1%) of the 24 368 participants who did not develop lung cancer in the year following the first follow-up screening round could have safely skipped it, at the expense of delayed diagnosis of 0 (0.0%, 0.0% to 2.7%), 8 (4.6%, 2.2% to 9.2%), 17 (9.8%, 6.0% to 15.4%), 44 (25.3%, 19.2% to 32.5%) and 70 (40.2%, 33.0% to 47.9%) of the 174 lung cancers, respectively.ConclusionsThe polynomial model, using both patient characteristics and baseline scan morphology, was significantly superior in assigning participants to 1-year or 2-year screening intervals. Implementing personalised follow-up intervals would enable hundreds of participants to skip a screening round per lung cancer diagnosis delayed.

Contains fulltext : 196781.pdf (Publisher’s version ) (Open Access) INTRODUCTION: Lung cancer presenting as subsolid nodule (SSN) can show slow growth, hence treating SSN is controversial. Our aim was to determine the long-term outcome of subjects with unresected SSNs in lung cancer screening. METHODS: Since 2005, the Multicenter Italian Lung Detection (MILD) screening trial implemented active surveillance for persistent SSN, as opposed to early resection. Presence of SSNs was related to diagnosis of cancer at the site of SSN, elsewhere in the lung, or in the body. The risk of overall mortality and lung cancer mortality was tested by Cox proportional hazards model. RESULTS: SSNs were found in 16.9% (389 of 2303) of screenees. During 9.3 +/- 1.2 years of follow-up, the hazard ratio of lung cancer diagnosis in subjects with SSN was 6.77 (95% confidence interval: 3.39-13.54), with 73% (22 of 30) of cancers not arising from SSN (median time to diagnosis 52 months from SSN). Lung cancer-specific mortality in subjects with SSN was significantly increased (hazard ratio = 3.80; 95% confidence interval: 1.24-11.65) compared to subjects without lung nodules. Lung cancer arising from SSN did not lead to death within the follow-up period. CONCLUSIONS: Subjects with SSN in the MILD cohort showed a high risk of developing lung cancer elsewhere in the lung, with only a minority of cases arising from SSN, and never representing the cause of death. These results show the safety of active surveillance for conservative management of SSN until signs of solid component growth and the need for prolonged follow-up because of high risk of other cancers.

Researchers from Google AI have presented results obtained using a deep learning model for the detection of lung cancer in screening CT images. The authors report a level of performance similar to, or better than, that of radiologists. However, these claims are currently too strong. The model is promising but needs further validation and could only be implemented if screening guidelines were adjusted to accept recommendations from black-box proprietary AI systems.

Purpose: Current computer-aided detection (CAD) systems for pulmonary nodules in computed tomography (CT) scans have a good performance for relatively small nodules, but often fail to detect the much rarer larger nodules, which are more likely to be cancerous. We present a novel CAD system specifically designed to detect solid nodules larger than 10 mm. Methods: The proposed detection pipeline is initiated by a three-dimensional lung segmentation algorithm optimized to include large nodules attached to the pleural wall via morphological processing. An additional preprocessing is used to mask out structures outside the pleural space to ensure that pleural and parenchymal nodules have a similar appearance. Next, nodule candidates are obtained via a multistage process of thresholding and morphological operations, to detect both larger and smaller candidates. After segmenting each candidate, a set of 24 features based on intensity, shape, blobness, and spatial context are computed. A radial basis support vector machine (SVM) classifier was used to classify nodule candidates, and performance was evaluated using ten-fold cross-validation on the full publicly available lung image database consortium database. Results: The proposed CAD system reaches a sensitivity of 98.3% (234/238) and 94.1% (224/238) large nodules at an average of 4.0 and 1.0 false positives/scan, respectively. Conclusions: The authors conclude that the proposed dedicated CAD system for large pulmonary nodules can identify the vast majority of highly suspicious lesions in thoracic CT scans with a small number of false positives.

Contains fulltext : 153905.pdf (Publisher’s version ) (Open Access) The malignancy of lung nodules is most often detected by analyzing changes of the nodule diameter in follow-up scans. A recent study showed that comparing the volume or the mass of a nodule over time is much more significant than comparing the diameter. Since the survival rate is higher when the disease is still in an early stage it is important to detect the growth rate as soon as possible. However manual segmentation of a volume is time-consuming. Whereas there are several well evaluated methods for the segmentation of solid nodules, less work is done on subsolid nodules which actually show a higher malignancy rate than solid nodules. In this work we present a fast, semi-automatic method for segmentation of subsolid nodules.As minimal user interaction the method expects a user-drawn stroke on the largest diameter of the nodule. First, a threshold-based region growing is performed based on intensity analysis of the nodule region and surrounding parenchyma. In the next step the chest wall is removed by a combination of a connected component analyses and convex hull calculation. Finally, attached vessels are detached by morphological operations.The method was evaluated on all nodules of the publicly available LIDC/IDRI database that were manually segmented and rated as non-solid or part-solid by four radiologists (Dataset 1) and three radiologists (Dataset 2). For these 59 nodules the Jaccard index for the agreement of the proposed method with the manual reference segmentations was 0.52/0.50 (Dataset 1/Dataset 2) compared to an inter-observer agreement of the manual segmentations of 0.54/0.58 (Dataset 1/Dataset 2). Furthermore, the inter-observer agreement using the proposed method (i.e. different input strokes) was analyzed and gave a Jaccard index of 0.74/0.74 (Dataset 1/Dataset 2).The presented method provides satisfactory segmentation results with minimal observer effort in minimal time and can reduce the inter-observer variability for segmentation of subsolid nodules in clinical routine.

Contains fulltext : 153206.pdf (Publisher’s version ) (Open Access) We aimed to test the interscan variation of semi-automatic volumetry of subsolid nodules (SSNs), as growth evaluation is important for SSN management.From a lung cancer screening trial all SSNs that were stable over at least 3 months were included (N = 44). SSNs were quantified on the baseline CT by two observers using semi-automatic volumetry software for effective diameter, volume, and mass. One observer also measured the SSNs on the second CT 3 months later. Interscan variation was evaluated using Bland-Altman plots. Observer agreement was calculated as intraclass correlation coefficient (ICC). Data are presented as mean (± standard deviation) or median and interquartile range (IQR). A Mann-Whitney U test was used for the analysis of the influence of adjustments on the measurements.Semi-automatic measurements were feasible in all 44 SSNs. The interscan limits of agreement ranged from -12.0 \% to 9.7 \% for diameter, -35.4 \% to 28.6 \% for volume and -27.6 \% to 30.8 \% for mass. Agreement between observers was good with intraclass correlation coefficients of 0.978, 0.957, and 0.968 for diameter, volume, and mass, respectively.Our data suggest that when using our software an increase in mass of 30 \% can be regarded as significant growth.• Recently, recommendations regarding subsolid nodules have stressed the importance of growth quantification. • Volumetric measurement of subsolid nodules is feasible with good interscan agreement. • Increase of mass of 30 \% can be regarded as significant growth.

Contains fulltext : 154320.pdf (Publisher’s version ) (Open Access) Pulmonary subsolid nodules (SSNs) have a high likelihood of malignancy, but are often indolent. A conservative treatment approach may therefore be suitable. The aim of the current study was to evaluate whether close follow-up of SSNs with computed tomography may be a safe approach. The study population consisted of participants of the Dutch-Belgian lung cancer screening trial (Nederlands Leuvens Longkanker Screenings Onderzoek; NELSON). All SSNs detected during the trial were included in this analysis. Retrospectively, all persistent SSNs and SSNs that were resected after first detection were segmented using dedicated software, and maximum diameter, volume and mass were measured. Mass doubling time (MDT) was calculated. In total 7135 volunteers were included in the current analysis. 264 (3.3\%) SSNs in 234 participants were detected during the trial. 147 (63\%) of these SSNs in 126 participants disappeared at follow-up, leaving 117 persistent or directly resected SSNs in 108 (1.5\%) participants available for analysis. The median follow-up time was 95 months (range 20-110). 33 (28\%) SSNs were resected and 28 of those were (pre-) invasive. None of the non-resected SSNs progressed into a clinically relevant malignancy. Persistent SSNs rarely developed into clinically manifest malignancies unexpectedly. Close follow-up with computed tomography may be a safe option to monitor changes.

We train and apply convolutional neural networks, a machine learning technique developed to learn from and classify image data, to Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) imaging for the identification of potential strong lensing systems. An ensemble of four convolutional neural networks was trained on images of simulated galaxy-galaxy lenses. The training sets consisted of a total of 62,406 simulated lenses and 64,673 non-lens negative examples generated with two different methodologies. The networks were able to learn the features of simulated lenses with accuracy of up to 99.8% and a purity and completeness of 94-100% on a test set of 2000 simulations. An ensemble of trained networks was applied to all of the 171 square degrees of the CFHTLS wide field image data, identifying 18,861 candidates including 63 known and 139 other potential lens candidates. A second search of 1.4 million early type galaxies selected from the survey catalog as potential deflectors, identified 2,465 candidates including 117 previously known lens candidates, 29 confirmed lenses/high-quality lens candidates, 266 novel probable or potential lenses and 2097 candidates we classify as false positives. For the catalog-based search we estimate a completeness of 21-28% with respect to detectable lenses and a purity of 15%, with a false-positive rate of 1 in 671 images tested. We predict a human astronomer reviewing candidates produced by the system would identify ~20 probable lenses and 100 possible lenses per hour in a sample selected by the robot. Convolutional neural networks are therefore a promising tool for use in the search for lenses in current and forthcoming surveys such as the Dark Energy Survey and the Large Synoptic Survey Telescope., Comment: 16 pages, 8 figures. Accepted by MNRAS

The introduction of lung cancer screening programs will produce an unprecedented amount of chest CT scans in the near future, which radiologists will have to read in order to decide on a patient follow-up strategy. According to the current guidelines, the workup of screen-detected nodules strongly relies on nodule size and nodule type. In this paper, we present a deep learning system based on multi-stream multi-scale convolutional networks, which automatically classifies all nodule types relevant for nodule workup. The system processes raw CT data containing a nodule without the need for any additional information such as nodule segmentation or nodule size and learns a representation of 3D data by analyzing an arbitrary number of 2D views of a given nodule. The deep learning system was trained with data from the Italian MILD screening trial and validated on an independent set of data from the Danish DLCST screening trial. We analyze the advantage of processing nodules at multiple scales with a multi-stream convolutional network architecture, and we show that the proposed deep learning system achieves performance at classifying nodule type that surpasses the one of classical machine learning approaches and is within the inter-observer variability among four experienced human observers.

The purpose of this study is to develop a computed tomography (CT) biomarker of emphysema that is robust across reconstruction settings, and evaluate its ability to predict mortality in patients at high risk for lung cancer. Data included baseline CT scans acquired between August 2002 and April 2004 from 1737 deceased subjects and 5740 surviving controls taken from the National Lung Screening Trial. Emphysema scores were computed in the original scans (origES) and after applying resampling, normalization and bullae analysis (normES). We compared the prognostic value of normES versus origES for lung cancer and all-cause mortality by computing the area under the receiver operator characteristic curve (AUC) and the net reclassification improvement (NRI) for follow-up times of 1-7 years. normES was a better predictor of mortality than origES. The 95% confidence intervals for the differences in AUC values indicated a significant difference for all-cause mortality for 2 through 6 years of follow-up, and for lung cancer mortality for 1 through 7 years of follow-up. 95% confidence intervals in NRI values showed a statistically significant improvement in classification for all-cause mortality for 2 through 7 years of follow-up, and for lung cancer mortality for 3 through 7 years of follow-up. Contrary to conventional emphysema score, our normalized emphysema score is a good predictor of all-cause and lung cancer mortality in settings where multiple CT scanners and protocols are used.

A convolutional network architecture is presented to determine bounding boxes around six organs in thoraxabdomen CT scans. A single network for each orthogonal view determines the presence of lungs, kidneys, spleen and liver. We show that an architecture that takes additional slices before and after the slice of interest as an additional input outperforms an architecture that processes single slices. From the slice-based analysis, a bounding box around the structures of interest can be computed. The system uses 6 convolutional, 4 pooling and one fully connected layer and uses 333 scans for training and 110 for validation. The test set contains 110 scans. The average Dice score of the proposed method was 0.95 and 0.95 for the lungs, 0.59 and 0.58 for the kidneys, 0.83 for the liver and 0.63 for the spleen. This paper shows that automatic localization of organs using multi-label convolution neural networks is possible. This architecture can likely be used to identify other organs of interest as well.

Contains fulltext : 179531.pdf (Publisher’s version ) (Open Access) Automatic detection of pulmonary nodules in thoracic computed tomography (CT) scans has been an active area of research for the last two decades. However, there have only been few studies that provide a comparative performance evaluation of different systems on a common database. We have therefore set up the LUNA16 challenge, an objective evaluation framework for automatic nodule detection algorithms using the largest publicly available reference database of chest CT scans, the LIDC-IDRI data set. In LUNA16, participants develop their algorithm and upload their predictions on 888 CT scans in one of the two tracks: 1) the complete nodule detection track where a complete CAD system should be developed, or 2) the false positive reduction track where a provided set of nodule candidates should be classified. This paper describes the setup of LUNA16 and presents the results of the challenge so far. Moreover, the impact of combining individual systems on the detection performance was also investigated. It was observed that the leading solutions employed convolutional networks and used the provided set of nodule candidates. The combination of these solutions achieved an excellent sensitivity of over 95% at fewer than 1.0 false positives per scan. This highlights the potential of combining algorithms to improve the detection performance. Our observer study with four expert readers has shown that the best system detects nodules that were missed by expert readers who originally annotated the LIDC-IDRI data. We released this set of additional nodules for further development of CAD systems. 13 p.

Contains fulltext : 174818.pdf (Publisher’s version ) (Open Access) Purpose To evaluate the added value of Lung CT Screening Reporting and Data System (Lung-RADS) assessment category 4X over categories 3, 4A, and 4B for differentiating between benign and malignant subsolid nodules (SSNs). Materials and Methods SSNs on all baseline computed tomographic (CT) scans from the National Lung Cancer Trial that would have been classified as Lung-RADS category 3 or higher were identified, resulting in 374 SSNs for analysis. An experienced screening radiologist volumetrically segmented all solid cores and located all malignant SSNs visible on baseline scans. Six experienced chest radiologists independently determined which nodules to upgrade to category 4X, a recently introduced category for lesions that demonstrate additional features or imaging findings that increase the suspicion of malignancy. Malignancy rates of purely size-based categories and category 4X were compared. Furthermore, the false-positive rates of category 4X lesions were calculated and observer variability was assessed by using Fleiss kappa statistics. Results The observers upgraded 15%-24% of the SSNs to category 4X. The malignancy rate for 4X nodules varied from 46% to 57% per observer and was substantially higher than the malignancy rates of categories 3, 4A, and 4B SSNs without observer intervention (9%, 19%, and 23%, respectively). On average, the false-positive rate for category 4X nodules was 7% for category 3 SSNs, 7% for category 4A SSNs, and 19% for category 4B SSNs. Of the falsely upgraded benign lesions, on average 27% were transient. The agreement among the observers was moderate, with an average kappa value of 0.535 (95% confidence interval: 0.509, 0.561). Conclusion The inclusion of a 4X assessment category for lesions suspicious for malignancy in a nodule management tool is of added value and results in high malignancy rates in the hands of experienced radiologists. Proof of the transient character of category 4X lesions at short-term follow-up could avoid unnecessary invasive management. (c) RSNA, 2017.

The introduction of lung cancer screening programs will produce an unprecedented amount of chest CT scans in the near future, which radiologists will have to read in order to decide on a patient follow-up strategy. According to the current guidelines, the workup of screen-detected nodules strongly relies on nodule size and nodule type. In this paper, we present a deep learning system based on multi-stream multi-scale convolutional networks, which automatically classifies all nodule types relevant for nodule workup. The system processes raw CT data containing a nodule without the need for any additional information such as nodule segmentation or nodule size and learns a representation of 3D data by analyzing an arbitrary number of 2D views of a given nodule. The deep learning system was trained with data from the Italian MILD screening trial and validated on an independent set of data from the Danish DLCST screening trial. We analyze the advantage of processing nodules at multiple scales with a multi-stream convolutional network architecture, and we show that the proposed deep learning system achieves performance at classifying nodule type that surpasses the one of classical machine learning approaches and is within the inter-observer variability among four experienced human observers., Published on Scientific Reports

Contains fulltext : 154962.pdf (Publisher’s version ) (Open Access) To determine whether semiautomatic volumetric software can differentiate part-solid from nonsolid pulmonary nodules and aid quantification of the solid component.As per reference standard, 115 nodules were differentiated into nonsolid and part-solid by two radiologists; disagreements were adjudicated by a third radiologist. The diameters of solid components were measured manually. Semiautomatic volumetric measurements were used to identify and quantify a possible solid component, using different Hounsfield unit (HU) thresholds. The measurements were compared with the reference standard and manual measurements.The reference standard detected a solid component in 86 nodules. Diagnosis of a solid component by semiautomatic software depended on the threshold chosen. A threshold of -300 HU resulted in the detection of a solid component in 75 nodules with good sensitivity (90 \%) and specificity (88 \%). At a threshold of -130 HU, semiautomatic measurements of the diameter of the solid component (mean 2.4 mm, SD 2.7 mm) were comparable to manual measurements at the mediastinal window setting (mean 2.3 mm, SD 2.5 mm [p = 0.63]).Semiautomatic segmentation of subsolid nodules could diagnose part-solid nodules and quantify the solid component similar to human observers. Performance depends on the attenuation segmentation thresholds. This method may prove useful in managing subsolid nodules.• Semiautomatic segmentation can accurately differentiate nonsolid from part-solid pulmonary nodules • Semiautomatic segmentation can quantify the solid component similar to manual measurements • Semiautomatic segmentation may aid management of subsolid nodules following Fleischner Society recommendations • Performance for the segmentation of subsolid nodules depends on the chosen attenuation thresholds.

Contains fulltext : 118196.pdf (Publisher’s version ) (Open Access) OBJECTIVE. The purpose of this study was to investigate a new software program for semiautomatic measurement of the volume and mass of ground-glass nodules (GGNs) in a chest phantom and to investigate the influence of CT scanner, reconstruction filter, tube voltage, and tube current. MATERIALS AND METHODS. We used an anthropomorphic chest phantom with eight artificial GGNs with two different CT attenuations and four different volumes. CT scans were obtained with four models of CT scanner at 120 kVp and 25 mAs with a soft and a sharp reconstruction filter. On the 256-MDCT scanner, the tube current-exposure time product and tube voltage settings were varied. GGNs were measured with software that automatically segmented the nodules. Absolute percentage error (APE) was calculated for volume, mass, and density. Wilcoxon signed rank, Mann-Whitney U, and Kruskal-Wallis tests were used for analysis. RESULTS. Volume and mass did not differ significantly from the true values. When measurements were expressed as APE, the error range was 2-36\% for volume and 5-46\% for mass, which was significantly different from no error. We did not find significant differences in APE between CT scanners with filters for lower tube current for volume or lower tube voltage for mass. CONCLUSION. Computer-aided segmentation and mass and volume measurements of GGNs with the prototype software had promising results in this study.

Contains fulltext : 181940.pdf (Publisher’s version ) (Open Access) PURPOSE: Lung-RADS proposes malignancy probabilities for categories 2 (15%). The purpose of this study was to quantify and compare malignancy rates for Lung-RADS 2 and 4B subsolid nodules (SSNs) on a nodule base. METHODS: We identified all baseline SSNs eligible for Lung-RADS 2 and 4B in the National Lung Screening Trial (NLST) database. Solid cores and nodule locations were annotated using in-house software. Malignant SSNs were identified by an experienced radiologist using NLST information. Malignancy rates and percentages of persistence were calculated. RESULTS: Of the Lung-RADS 2SSNs, 94.3% (1790/1897) could be located on chest CTs. Likewise, 95.1% (331/348) of part-solid nodules >/=6 mm in diameter could be located. Of these, 120 had a solid core >/=8 mm, corresponding to category 4B. Category 2 SSNs showed a malignancy rate of 2.5%, exceeding slightly the proposed rate of

Contains fulltext : 164462.pdf (Publisher’s version ) (Open Access) We propose a novel Computer-Aided Detection (CAD) system for pulmonary nodules using multi-view convolutional networks (ConvNets), for which discriminative features are automatically learnt from the training data. The network is fed with nodule candidates obtained by combining three candidate detectors specifically designed for solid, subsolid, and large nodules. For each candidate, a set of 2-D patches from differently oriented planes is extracted. The proposed architecture comprises multiple streams of 2-D ConvNets, for which the outputs are combined using a dedicated fusion method to get the final classification. Data augmentation and dropout are applied to avoid overfitting. On 888 scans of the publicly available LIDCIDRI dataset, our method reaches high detection sensitivities of 85.4% and 90.1% at 1 and 4 false positives per scan, respectively. An additional evaluation on independent datasets from the ANODE09 challenge and DLCST is performed. We showed that the proposed multi-view ConvNets is highly suited to be used for false positive reduction of a CAD system.

Contains fulltext : 172195.pdf (Publisher’s version ) (Open Access) To implement a cost-effective low-dose computed tomography (LDCT) lung cancer screening program at the population level, accurate and efficient interpretation of a large volume of LDCT scans is needed. The objective of this study was to evaluate a workflow strategy to identify abnormal LDCT scans in which a technician assisted by computer vision (CV) software acts as a first reader with the aim to improve speed, consistency, and quality of scan interpretation.Without knowledge of the diagnosis, a technician reviewed 828 randomly batched scans (136 with lung cancers, 556 with benign nodules, and 136 without nodules) from the baseline Pan-Canadian Early Detection of Lung Cancer Study that had been annotated by the CV software CIRRUS Lung Screening (Diagnostic Image Analysis Group, Nijmegen, The Netherlands). The scans were classified as either normal (no nodules \geq1 mm or benign nodules) or abnormal (nodules or other abnormality). The results were compared with the diagnostic interpretation by Pan-Canadian Early Detection of Lung Cancer Study radiologists.The overall sensitivity and specificity of the technician in identifying an abnormal scan were 97.8\% (95\% confidence interval: 96.4-98.8) and 98.0\% (95\% confidence interval: 89.5-99.7), respectively. Of the 112 prevalent nodules that were found to be malignant in follow-up, 92.9\% were correctly identified by the technician plus CV compared with 84.8\% by the study radiologists. The average time taken by the technician to review a scan after CV processing was 208 ± 120 seconds.Prescreening CV software and a technician as first reader is a promising strategy for improving the consistency and quality of screening interpretation of LDCT scans.