•Vertebrae are segmented with an iterative instance segmentation algorithm.•The method does not make assumptions about the number of visible vertebrae.•Detected vertebrae are anatomically labeled ...using a global probabilistic model.•A fully convolutional neural network performs both segmentation and identification.•Vertebra segmentations and identifications are evaluated on five CT and MR datasets.
Display omitted
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.
Intracranial internal carotid artery (iICA) calcification is associated with stroke and is often seen as a proxy of atherosclerosis of the intima. However, it was recently shown that these ...calcifications are predominantly located in the tunica media and internal elastic lamina (medial calcification). Intimal and medial calcifications are thought to have a different pathogenesis and clinical consequences and can only be distinguished through ex vivo histological analysis. Therefore, our aim was to develop CT scoring method to distinguish intimal and medial iICA calcification in vivo.
First, in both iICAs of 16 cerebral autopsy patients the intimal and/or medial calcification area was histologically assessed (142 slides). Brain CT images of these patients were matched to the corresponding histological slides to develop a CT score that determines intimal or medial calcification dominance. Second, performance of the CT score was assessed in these 16 patients. Third, reproducibility was tested in a separate cohort.
First, CT features of the score were circularity (absent, dot(s), <90°, 90-270° or 270-360°), thickness (absent, ≥1.5mm, or <1.5mm), and morphology (indistinguishable, irregular/patchy or continuous). A high sum of features represented medial and a lower sum intimal calcifications. Second, in the 16 patients the concordance between the CT score and the dominant calcification type was reasonable. Third, the score showed good reproducibility (kappa: 0.72 proportion of agreement: 0.82) between the categories intimal, medial or absent/indistinguishable.
The developed CT score shows good reproducibility and can differentiate reasonably well between intimal and medial calcification dominance in the iICA, allowing for further (epidemiological) studies on iICA calcification.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Objectives
Opportunistic screening for osteoporosis using computed tomography (CT) examinations that happen to visualise the spine can be used to identify patients with osteoporosis. We sought to ...verify the diagnostic performance of vertebral Hounsfield unit (HU) measurements on routine CT examinations for diagnosing osteoporosis in a separate, external population.
Methods
Consecutive patients who underwent a CT examination of the chest or abdomen and had also received a dual-energy X-ray absorptiometry (DXA) test were retrospectively included. CTs were evaluated for vertebral fractures and vertebral attenuation (density) values were measured. Diagnostic performance measures and the area under the receiver operator characteristics curve (AUC) for diagnosing osteoporosis were calculated.
Results
Three hundred and two patients with a mean age of 57.9 years were included, of which 82 (27 %) had osteoporosis according to DXA and 65 (22 %) had vertebral fractures. The diagnostic performance for vertebral HU measurements was modest, with a maximal AUC of 0.74 (0.68 – 0.80). At that optimal threshold the sensitivity was 62 % (51 – 72 %) and the specificity was 79 % (74 – 84 %).
Conclusions
We confirmed that simple trabecular vertebral density measurements on routine CT contain diagnostic information related to bone mineral density as measured by DXA, albeit with substantially lower diagnostic accuracy than previously reported.
Key Points
•
We externally validated the value of vertebral trabecular bone attenuation for osteoporosis
•
These diagnostic performance measures were, however, substantially lower than previously reported
•
This information might be useful when considering the implementation of opportunistic osteoporosis screening
Summary Background Low-dose CT screening is recommended for individuals at high risk of developing lung cancer. However, CT screening does not detect all lung cancers: some might be missed at ...screening, and others can develop in the interval between screens. The NELSON trial is a randomised trial to assess the effect of screening with increasing screening intervals on lung cancer mortality. In this prespecified analysis, we aimed to assess screening test performance, and the epidemiological, radiological, and clinical characteristics of interval cancers in NELSON trial participants assigned to the screening group. Methods Eligible participants in the NELSON trial were those aged 50–75 years, who had smoked 15 or more cigarettes per day for more than 25 years or ten or more cigarettes for more than 30 years, and were still smoking or had quit less than 10 years ago. We included all participants assigned to the screening group who had attended at least one round of screening. Screening test results were based on volumetry using a two-step approach. Initially, screening test results were classified as negative, indeterminate, or positive based on nodule presence and volume. Subsequently, participants with an initial indeterminate result underwent follow-up screening to classify their final screening test result as negative or positive, based on nodule volume doubling time. We obtained information about all lung cancer diagnoses made during the first three rounds of screening, plus an additional 2 years of follow-up from the national cancer registry. We determined epidemiological, radiological, participant, and tumour characteristics by reassessing medical files, screening CTs, and clinical CTs. The NELSON trial is registered at www.trialregister.nl , number ISRCTN63545820. Findings 15 822 participants were enrolled in the NELSON trial, of whom 7915 were assigned to low-dose CT screening with increasing interval between screens, and 7907 to no screening. We included 7155 participants in our study, with median follow-up of 8·16 years (IQR 7·56–8·56). 187 (3%) of 7155 screened participants were diagnosed with 196 screen-detected lung cancers, and another 34 (<1%; 19 56% in the first year after screening, and 15 44% in the second year after screening) were diagnosed with 35 interval cancers. For the three screening rounds combined, with a 2-year follow-up, sensitivity was 84·6% (95% CI 79·6–89·2), specificity was 98·6% (95% CI 98·5–98·8), positive predictive value was 40·4% (95% CI 35·9–44·7), and negative predictive value was 99·8% (95% CI 99·8–99·9). Retrospective assessment of the last screening CT and clinical CT in 34 patients with interval cancer showed that interval cancers were not visible in 12 (35%) cases. In the remaining cases, cancers were visible when retrospectively assessed, but were not diagnosed because of radiological detection and interpretation errors (17 50%), misclassification by the protocol (two 6%), participant non-compliance (two 6%), and non-adherence to protocol (one 3%). Compared with screen-detected cancers, interval cancers were diagnosed at more advanced stages (29 83% of 35 interval cancers vs 44 22% of 196 screen-detected cancers diagnosed in stage III or IV; p<0·0001), were more often small-cell carcinomas (seven 20% vs eight 4%; p=0·003) and less often adenocarcinomas (nine 26% vs 102 52%; p=0·005). Interpretation Lung cancer screening in the NELSON trial yielded high specificity and sensitivity, with only a small number of interval cancers. The results of this study could be used to improve screening algorithms, and reduce the number of missed cancers. Funding Zorgonderzoek Nederland Medische Wetenschappen and Koningin Wilhelmina Fonds.
•Peri-fissural nodules (PFNs) have been proven to be bening nodules, for which no follow-up is needed.•Automatic classsification of PFNs would make lung cancer screening more efficient and reduce the ...number of follow-up.•Automatic classsification of PFNs would make lung cancer screening more efficient and reduce the number of follow-up.•State-of-the-art machine learning techniques approach human performance in PFNs classification.
Display omitted
In this paper, we tackle the problem of automatic classification of pulmonary peri-fissural nodules (PFNs). The classification problem is formulated as a machine learning approach, where detected nodule candidates are classified as PFNs or non-PFNs. Supervised learning is used, where a classifier is trained to label the detected nodule. The classification of the nodule in 3D is formulated as an ensemble of classifiers trained to recognize PFNs based on 2D views of the nodule. In order to describe nodule morphology in 2D views, we use the output of a pre-trained convolutional neural network known as OverFeat. We compare our approach with a recently presented descriptor of pulmonary nodule morphology, namely Bag of Frequencies, and illustrate the advantages offered by the two strategies, achieving performance of AUC = 0.868, which is close to the one of human experts.
In the USA annual lung cancer screening is recommended. However, the optimal screening strategy (eg, screening interval, screening rounds) is unknown. This study provides results of the fourth ...screening round after a 2.5-year interval in the Dutch-Belgian Lung Cancer Screening trial (NELSON).
Europe's largest, sufficiently powered randomised lung cancer screening trial was designed to determine whether low-dose CT screening reduces lung cancer mortality by ≥25% compared with no screening after 10 years of follow-up. The screening arm (n=7915) received screening at baseline, after 1 year, 2 years and 2.5 years. Performance of the NELSON screening strategy in the final fourth round was evaluated. Comparisons were made between lung cancers detected in the first three rounds, in the final round and during the 2.5-year interval.
In round 4, 46 cancers were screen-detected and there were 28 interval cancers between the third and fourth screenings. Compared with the second round screening (1-year interval), in round 4 a higher proportion of stage IIIb/IV cancers (17.3% vs 6.8%, p=0.02) and higher proportions of squamous-cell, bronchoalveolar and small-cell carcinomas (p=0.001) were detected. Compared with a 2-year interval, the 2.5-year interval showed a higher non-significant stage distribution (stage IIIb/IV 17.3% vs 5.2%, p=0.10). Additionally, more interval cancers manifested in the 2.5-year interval than in the intervals of previous rounds (28 vs 5 and 28 vs 19).
A 2.5-year interval reduced the effect of screening: the interval cancer rate was higher compared with the 1-year and 2-year intervals, and proportion of advanced disease stage in the final round was higher compared with the previous rounds.
ISRCTN63545820.
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 the 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. The 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 convolutional neural networks: 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 into the regions that are contributed to the calcium score. Experiments were performed using 903 cardiac CT and 1687 chest CT scans. The method predicted calcium scores in less than 0.3 s. The 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 the 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 the 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.
Objectives
To present the results of a systematic literature search aimed at determining to what extent the radiation dose can be reduced with iterative reconstruction (IR) for cardiopulmonary and ...body imaging with computed tomography (CT) in the clinical setting and what the effects on image quality are with IR versus filtered back-projection (FBP) and to provide recommendations for future research on IR.
Methods
We searched Medline and Embase from January 2006 to January 2012 and included original research papers concerning IR for CT.
Results
The systematic search yielded 380 articles. Forty-nine relevant studies were included. These studies concerned: the chest(
n
= 26), abdomen(
n
= 16), both chest and abdomen(
n
= 1), head(
n
= 4), spine(
n
= 1), and no specific area (
n
= 1). IR reduced noise and artefacts, and it improved subjective and objective image quality compared to FBP at the same dose. Conversely, low-dose IR and normal-dose FBP showed similar noise, artefacts, and subjective and objective image quality. Reported dose reductions ranged from 23 to 76 % compared to locally used default FBP settings. However, IR has not yet been investigated for ultra-low-dose acquisitions with clinical diagnosis and accuracy as endpoints.
Conclusion
Benefits of IR include improved subjective and objective image quality as well as radiation dose reduction while preserving image quality. Future studies need to address the value of IR in ultra-low-dose CT with clinically relevant endpoints.
Key Points
•
Iterative reconstruction improves image quality of CT images at equal acquisition parameters.
•
IR preserves image quality compared to normal-dose filtered back-projection.
•
The reduced radiation dose made possible by IR is advantageous for patients.
•
IR has not yet been investigated with clinical diagnosis and accuracy as endpoints.
Summary Background The main challenge in CT screening for lung cancer is the high prevalence of pulmonary nodules and the relatively low incidence of lung cancer. Management protocols use thresholds ...for nodule size and growth rate to determine which nodules require additional diagnostic procedures, but these should be based on individuals' probabilities of developing lung cancer. In this prespecified analysis, using data from the NELSON CT screening trial, we aimed to quantify how nodule diameter, volume, and volume doubling time affect the probability of developing lung cancer within 2 years of a CT scan, and to propose and evaluate thresholds for management protocols. Methods Eligible participants in the NELSON trial were those aged 50–75 years, who have smoked 15 cigarettes or more per day for more than 25 years, or ten cigarettes or more for more than 30 years and were still smoking, or had stopped smoking less than 10 years ago. Participants were randomly assigned to low-dose CT screening at increasing intervals, or no screening. We included all participants assigned to the screening group who had attended at least one round of screening, and whose results were available from the national cancer registry database. We calculated lung cancer probabilities, stratified by nodule diameter, volume, and volume doubling time and did logistic regression analysis using diameter, volume, volume doubling time, and multinodularity as potential predictor variables. We assessed management strategies based on nodule threshold characteristics for specificity and sensitivity, and compared them to the American College of Chest Physicians (ACCP) guidelines. The NELSON trial is registered at www.trialregister.nl , number ISRCTN63545820. Findings Volume, volume doubling time, and volumetry-based diameter of 9681 non-calcified nodules detected by CT screening in 7155 participants in the screening group of NELSON were used to quantify lung cancer probability. Lung cancer probability was low in participants with a nodule volume of 100 mm3 or smaller (0·6% 95% CI 0·4–0·8) or maximum transverse diameter smaller than 5 mm (0·4% 0·2–0·7), and not significantly different from participants without nodules (0·4% 0·3–0·6, p=0·17 and p=1·00, respectively). Lung cancer probability was intermediate (requiring follow-up CT) if nodules had a volume of 100–300 mm3 (2·4% 95% CI 1·7–3·5) or a diameter 5–10 mm (1·3% 1·0–1·8). Volume doubling time further stratified the probabilities: 0·8% (95% CI 0·4–1·7) for volume doubling times 600 days or more, 4·0% (1·8–8·3) for volume doubling times 400–600 days, and 9·9% (6·9–14·1) for volume doubling times of 400 days or fewer. Lung cancer probability was high for participants with nodule volumes 300 mm3 or bigger (16·9% 95% CI 14·1–20·0) or diameters 10 mm or bigger (15·2% 12·7–18·1). The simulated ACCP management protocol yielded a sensitivity and specificity of 90·9% (95% CI 81·2–96·1), and 87·2% (86·4–87·9), respectively. A diameter-based protocol with volumetry-based nodule diameter yielded a higher sensitivity (92·4% 95% CI 83·1–97·1), and a higher specificity (90·0% 89·3–90·7). A volume-based protocol (with thresholds based on lung cancer probability) yielded the same sensitivity as the ACCP protocol (90·9% 95% CI 81·2–96·1), and a higher specificity (94·9% 94·4–95·4). Interpretation Small nodules (those with a volume <100 mm3 or diameter <5 mm) are not predictive for lung cancer. Immediate diagnostic evaluation is necessary for large nodules (≥300 mm3 or ≥10 mm). Volume doubling time assessment is advocated only for intermediate-sized nodules (with a volume ranging between 100–300 mm3 or diameter of 5–10 mm). Nodule management protocols based on these thresholds performed better than the simulated ACCP nodule protocol. Funding Zorgonderzoek Nederland Medische Wetenschappen and Koningin Wilhelmina Fonds.
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 F 1 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 F 1 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.
PDF
