TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images.

Purpose: To present a deep learning segmentation model that can automatically and robustly segment all major anatomic structures on body CT images. Materials and Methods: In this retrospective study, 1204 CT examinations (from 2012, 2016, and 2020) were used to segment 104 anatomic structures (27 organs, 59 bones, 10 muscles, and eight vessels) relevant for use cases such as organ volumetry, disease characterization, and surgical or radiation therapy planning. The CT images were randomly sampled from routine clinical studies and thus represent a real-world dataset (different ages, abnormalities, scanners, body parts, sequences, and sites). The authors trained an nnU-Net segmentation algorithm on this dataset and calculated Dice similarity coefficients to evaluate the model's performance. The trained algorithm was applied to a second dataset of 4004 whole-body CT examinations to investigate age-dependent volume and attenuation changes. Results: The proposed model showed a high Dice score (0.943) on the test set, which included a wide range of clinical data with major abnormalities. The model significantly outperformed another publicly available segmentation model on a separate dataset (Dice score, 0.932 vs 0.871; P < .001). The aging study demonstrated significant correlations between age and volume and mean attenuation for a variety of organ groups (eg, age and aortic volume [rs = 0.64; P < .001]; age and mean attenuation of the autochthonous dorsal musculature [rs = -0.74; P < .001]). Conclusion: The developed model enables robust and accurate segmentation of 104 anatomic structures. The annotated dataset (https://doi.org/10.5281/zenodo.6802613) and toolkit (https://www.github.com/wasserth/TotalSegmentator) are publicly available.Keywords: CT, Segmentation, Neural Networks Supplemental material is available for this article. © RSNA, 2023See also commentary by Sebro and Mongan in this issue.

Impact of retraining a deep learning algorithm for improving guideline-compliant aortic diameter measurements on non-gated chest CT.

PURPOSE/OBJECTIVE: Reliable detection of thoracic aortic dilatation (TAD) is mandatory in clinical routine. For ECG-gated CT angiography, automated deep learning (DL) algorithms are established for diameter measurements according to current guidelines. For non-ECG gated CT (contrast enhanced (CE) and non-CE), however, only a few reports are available. In these reports, classification as TAD is frequently unreliable with variable result quality depending on anatomic location with the aortic root presenting with the worst results. Therefore, this study aimed to explore the impact of re-training on a previously evaluated DL tool for aortic measurements in a cohort of non-ECG gated exams. METHODS & MATERIALS: A cohort of 995 patients (68 ± 12 years) with CE (n = 392) and non-CE (n = 603) chest CT exams was selected which were classified as TAD by the initial DL tool. The re-trained version featured improved robustness of centerline fitting and cross-sectional plane placement. All cases were processed by the re-trained DL tool version. DL results were evaluated by a radiologist regarding plane placement and diameter measurements. Measurements were classified as correctly measured diameters at each location whereas false measurements consisted of over-/under-estimation of diameters. RESULTS: We evaluated 8948 measurements in 995 exams. The re-trained version performed 8539/8948 (95.5%) of diameter measurements correctly. 3765/8948 (42.1%) of measurements were correct in both versions, initial and re-trained DL tool (best: distal arch 655/995 (66%), worst: Aortic sinus (AS) 221/995 (22%)). In contrast, 4456/8948 (49.8%) measurements were correctly measured only by the re-trained version, in particular at the aortic root (AS: 564/995 (57%), sinotubular junction: 697/995 (70%)). In addition, the re-trained version performed 318 (3.6%) measurements which were not available previously. A total of 228 (2.5%) cases showed false measurements because of tilted planes and 181 (2.0%) over-/under-segmentations with a focus at AS (n = 137 (14%) and n = 73 (7%), respectively). CONCLUSION: Re-training of the DL tool improved diameter assessment, resulting in a total of 95.5% correct measurements. Our data suggests that the re-trained DL tool can be applied even in non-ECG-gated chest CT including both, CE and non-CE exams.

Reduction in Radiologist Interpretation Time of Serial CT and MR Imaging Findings with Deep Learning Identification of Relevant Priors, Series and Finding Locations.

RATIONALE AND OBJECTIVES: Finding comparison to relevant prior studies is a requisite component of the radiology workflow. The purpose of this study was to evaluate the impact of a deep learning tool simplifying this time-consuming task by automatically identifying and displaying the finding in relevant prior studies. MATERIALS AND METHODS: The algorithm pipeline used in this retrospective study, TimeLens (TL), is based on natural language processing and descriptor-based image-matching algorithms. The dataset used for testing comprised 3872 series of 246 radiology examinations from 75 patients (189 CTs, 95 MRIs). To ensure a comprehensive testing, five finding types frequently encountered in radiology practice were included: aortic aneurysm, intracranial aneurysm, kidney lesion, meningioma, and pulmonary nodule. After a standardized training session, nine radiologists from three university hospitals performed two reading sessions on a cloud-based evaluation platform resembling a standard RIS/PACS. The task was to measure the diameter of the finding-of-interest on two or more exams (a most recent and at least one prior exam): first without use of TL, and a second session at an interval of at least 21 days with the use of TL. All user actions were logged for each round, including time needed to measure the finding at all timepoints, number of mouse clicks, and mouse distance traveled. The effect of TL was evaluated in total, per finding type, per reader, per experience (resident vs. board-certified radiologist), and per modality. Mouse movement patterns were analyzed with heatmaps. To assess the effect of habituation to the cases, a third round of readings was performed without TL. RESULTS: Across scenarios, TL reduced the average time needed to assess a finding at all timepoints by 40.1% (107 vs. 65 seconds; p < 0.001). Largest accelerations were demonstrated for assessment of pulmonary nodules (-47.0%; p < 0.001). Less mouse clicks (-17.2%) were needed for finding evaluation with TL, and mouse distance traveled was reduced by 38.0%. Time needed to assess the findings increased from round 2 to round 3 (+27.6%; p < 0.001). Readers were able to measure a given finding in 94.4% of cases on the series initially proposed by TL as most relevant series for comparison. The heatmaps showed consistently simplified mouse movement patterns with TL. CONCLUSION: A deep learning tool significantly reduced both the amount of user interactions with the radiology image viewer and the time needed to assess findings of interest on cross-sectional imaging with relevant prior exams.

Performance of a deep learning tool to detect missed aortic dilatation in a large chest CT cohort.

Purpose: Thoracic aortic (TA) dilatation (TAD) is a risk factor for acute aortic syndrome and must therefore be reported in every CT report. However, the complex anatomy of the thoracic aorta impedes TAD detection. We investigated the performance of a deep learning (DL) prototype as a secondary reading tool built to measure TA diameters in a large-scale cohort. Material and methods: Consecutive contrast-enhanced (CE) and non-CE chest CT exams with "normal" TA diameters according to their radiology reports were included. The DL-prototype (AIRad, Siemens Healthineers, Germany) measured the TA at nine locations according to AHA guidelines. Dilatation was defined as >45 mm at aortic sinus, sinotubular junction (STJ), ascending aorta (AA) and proximal arch and >40 mm from mid arch to abdominal aorta. A cardiovascular radiologist reviewed all cases with TAD according to AIRad. Multivariable logistic regression (MLR) was used to identify factors (demographics and scan parameters) associated with TAD classification by AIRad. Results: 18,243 CT scans (45.7% female) were successfully analyzed by AIRad. Mean age was 62.3 ± 15.9 years and 12,092 (66.3%) were CE scans. AIRad confirmed normal diameters in 17,239 exams (94.5%) and reported TAD in 1,004/18,243 exams (5.5%). Review confirmed TAD classification in 452/1,004 exams (45.0%, 2.5% total), 552 cases were false-positive but identification was easily possible using visual outputs by AIRad. MLR revealed that the following factors were significantly associated with correct TAD classification by AIRad: TAD reported at AA [odds ratio (OR): 1.12, p < 0.001] and STJ (OR: 1.09, p = 0.002), TAD found at >1 location (OR: 1.42, p = 0.008), in CE exams (OR: 2.1-3.1, p < 0.05), men (OR: 2.4, p = 0.003) and patients presenting with higher BMI (OR: 1.05, p = 0.01). Overall, 17,691/18,243 (97.0%) exams were correctly classified. Conclusions: AIRad correctly assessed the presence or absence of TAD in 17,691 exams (97%), including 452 cases with previously missed TAD independent from contrast protocol. These findings suggest its usefulness as a secondary reading tool by improving report quality and efficiency.

Automated quantification of airway wall thickness on chest CT using retina U-Nets - Performance evaluation and application to a large cohort of chest CTs of COPD patients.

PURPOSE: Airway wall thickening is a consequence of chronic inflammatory processes and usually only qualitatively described in CT radiology reports. The purpose of this study is to automatically quantify airway wall thickness in multiple airway generations and assess the diagnostic potential of this parameter in a large cohort of patients with Chronic Obstructive Pulmonary Disease (COPD). MATERIALS AND METHODS: This retrospective, single-center study included a series of unenhanced chest CTs. Inclusion criteria were the mentioning of an explicit COPD GOLD stage in the written radiology report and time period (01/2019-12/2021). A control group included chest CTs with completely unremarkable lungs according to the report. The DICOM images of all cases (axial orientation; slice-thickness: 1 mm; soft-tissue kernel) were processed by an AI algorithm pipeline consisting of (A) a 3D-U-Net for det detection and tracing of the bronchial tree centerlines (B) extraction of image patches perpendicular to the centerlines of the bronchi, and (C) a 2D U-Net for segmentation of airway walls on those patches. The performance of centerline detection and wall segmentation was assessed. The imaging parameter average wall thickness was calculated for bronchus generations 3-8 (AWT3-8) across the lungs. Mean AWT3-8 was compared between five groups (control, COPD Gold I-IV) using non-parametric statistics. Furthermore, the established emphysema score %LAV-950 was calculated and used to classify scans (normal vs. COPD) alone and in combination with AWT3-8. RESULTS: A total of 575 chest CTs were processed. Algorithm performance was very good (airway centerline detection sensitivity: 86.9%; airway wall segmentation Dice score: 0.86). AWT3-8 was statistically significantly greater in COPD patients compared to controls (2.03 vs. 1.87 mm, p < 0.001) and increased with COPD stage. The classifier that combined %LAV-950 and AWT3-8 was superior to the classifier using only %LAV-950 (AUC = 0.92 vs. 0.79). CONCLUSION: Airway wall thickness increases in patients suffering from COPD and is automatically quantifiable. AWT3-8 could become a CT imaging parameter in COPD complementing the established emphysema biomarker %LAV-950. CLINICAL RELEVANCE STATEMENT: Quantitative measurements considering the complete visible bronchial tree instead of qualitative description could enhance radiology reports, allow for precise monitoring of disease progression and diagnosis of early stages of disease.

Acute Pulmonary Embolism in COVID-19: A Potential Connection between Venous Congestion and Thrombus Distribution.

Background: Vascular abnormalities, including venous congestion (VC) and pulmonary embolism (PE), have been recognized as frequent COVID-19 imaging patterns and proposed as severity markers. However, the underlying pathophysiological mechanisms remain unclear. In this study, we aimed to characterize the relationship between VC, PE distribution, and alveolar opacities (AO). Methods: This multicenter observational registry (clinicaltrials.gov identifier NCT04824313) included 268 patients diagnosed with SARS-CoV-2 infection and subjected to contrast-enhanced CT between March and June 2020. Acute PE was diagnosed in 61 (22.8%) patients, including 17 females (27.9%), at a mean age of 61.7 ± 14.2 years. Demographic, laboratory, and outcome data were retrieved. We analyzed CT images at the segmental level regarding VC (qualitatively and quantitatively [diameter]), AO (semi-quantitatively as absent, <50%, or >50% involvement), clot location, and distribution related to VC and AO. Segments with vs. without PE were compared. Results: Out of 411 emboli, 82 (20%) were lobar or more proximal and 329 (80%) were segmental or subsegmental. Venous diameters were significantly higher in segments with AO (p = 0.031), unlike arteries (p = 0.138). At the segmental level, 77% of emboli were associated with VC. Overall, PE occurred in 28.2% of segments with AO vs. 21.8% without (p = 0.047). In the absence of VC, however, AO did not affect PE rates (p = 0.94). Conclusions: Vascular changes predominantly affected veins, and most PEs were located in segments with VC. In the absence of VC, AOs were not associated with the PE rate. VC might result from increased flow supported by the hypothesis of pulmonary arteriovenous anastomosis dysregulation as a relevant contributing factor.

Atri-U: assisted image analysis in routine cardiovascular magnetic resonance volumetry of the left atrium.

BACKGROUND: Artificial intelligence can assist in cardiac image interpretation. Here, we achieved a substantial reduction in time required to read a cardiovascular magnetic resonance (CMR) study to estimate left atrial volume without compromising accuracy or reliability. Rather than deploying a fully automatic black-box, we propose to incorporate the automated LA volumetry into a human-centric interactive image-analysis process. METHODS AND RESULTS: Atri-U, an automated data analysis pipeline for long-axis cardiac cine images, computes the atrial volume by: (i) detecting the end-systolic frame, (ii) outlining the endocardial borders of the LA, (iii) localizing the mitral annular hinge points and constructing the longitudinal atrial diameters, equivalent to the usual workup done by clinicians. In every step human interaction is possible, such that the results provided by the algorithm can be accepted, corrected, or re-done from scratch. Atri-U was trained and evaluated retrospectively on a sample of 300 patients and then applied to a consecutive clinical sample of 150 patients with various heart conditions. The agreement of the indexed LA volume between Atri-U and two experts was similar to the inter-rater agreement between clinicians (average overestimation of 0.8 mL/m2 with upper and lower limits of agreement of - 7.5 and 5.8 mL/m2, respectively). An expert cardiologist blinded to the origin of the annotations rated the outputs produced by Atri-U as acceptable in 97% of cases for step (i), 94% for step (ii) and 95% for step (iii), which was slightly lower than the acceptance rate of the outputs produced by a human expert radiologist in the same cases (92%, 100% and 100%, respectively). The assistance of Atri-U lead to an expected reduction in reading time of 66%-from 105 to 34 s, in our in-house clinical setting. CONCLUSIONS: Our proposal enables automated calculation of the maximum LA volume approaching human accuracy and precision. The optional user interaction is possible at each processing step. As such, the assisted process sped up the routine CMR workflow by providing accurate, precise, and validated measurement results.

Fully automated guideline-compliant diameter measurements of the thoracic aorta on ECG-gated CT angiography using deep learning.

BACKGROUND: Manually performed diameter measurements on ECG-gated CT-angiography (CTA) represent the gold standard for diagnosis of thoracic aortic dilatation. However, they are time-consuming and show high inter-reader variability. Therefore, we aimed to evaluate the accuracy of measurements of a deep learning-(DL)-algorithm in comparison to those of radiologists and evaluated measurement times (MT). METHODS: We retrospectively analyzed 405 ECG-gated CTA exams of 371 consecutive patients with suspected aortic dilatation between May 2010 and June 2019. The DL-algorithm prototype detected aortic landmarks (deep reinforcement learning) and segmented the lumen of the thoracic aorta (multi-layer convolutional neural network). It performed measurements according to AHA-guidelines and created visual outputs. Manual measurements were performed by radiologists using centerline technique. Human performance variability (HPV), MT and DL-performance were analyzed in a research setting using a linear mixed model based on 21 randomly selected, repeatedly measured cases. DL-algorithm results were then evaluated in a clinical setting using matched differences. If the differences were within 5 mm for all locations, the cases was regarded as coherent; if there was a discrepancy >5 mm at least at one location (incl. missing values), the case was completely reviewed. RESULTS: HPV ranged up to ±3.4 mm in repeated measurements under research conditions. In the clinical setting, 2,778/3,192 (87.0%) of DL-algorithm's measurements were coherent. Mean differences of paired measurements between DL-algorithm and radiologists at aortic sinus and ascending aorta were -0.45±5.52 and -0.02±3.36 mm. Detailed analysis revealed that measurements at the aortic root were over-/underestimated due to a tilted measurement plane. In total, calculated time saved by DL-algorithm was 3:10 minutes/case. CONCLUSIONS: The DL-algorithm provided coherent results to radiologists at almost 90% of measurement locations, while the majority of discrepent cases were located at the aortic root. In summary, the DL-algorithm assisted radiologists in performing AHA-compliant measurements by saving 50% of time per case.

Automated Detection of Pancreatic Cystic Lesions on CT Using Deep Learning.

Pancreatic cystic lesions (PCL) are a frequent and underreported incidental finding on CT scans and can transform into neoplasms with devastating consequences. We developed and evaluated an algorithm based on a two-step nnU-Net architecture for automated detection of PCL on CTs. A total of 543 cysts on 221 abdominal CTs were manually segmented in 3D by a radiology resident in consensus with a board-certified radiologist specialized in abdominal radiology. This information was used to train a two-step nnU-Net for detection with the performance assessed depending on lesions' volume and location in comparison to three human readers of varying experience. Mean sensitivity was 78.8 ± 0.1%. The sensitivity was highest for large lesions with 87.8% for cysts ≥220 mm3 and for lesions in the distal pancreas with up to 96.2%. The number of false-positive detections for cysts ≥220 mm3 was 0.1 per case. The algorithm's performance was comparable to human readers. To conclude, automated detection of PCL on CTs is feasible. The proposed model could serve radiologists as a second reading tool. All imaging data and code used in this study are freely available online.

Deep learning-based automated detection of pulmonary embolism on CT pulmonary angiograms: No significant effects on report communication times and patient turnaround in the emergency department nine months after technical implementation.

OBJECTIVES: Rapid communication of CT exams positive for pulmonary embolism (PE) is crucial for timely initiation of anticoagulation and patient outcome. It is unknown if deep learning automated detection of PE on CT Pulmonary Angiograms (CTPA) in combination with worklist prioritization and an electronic notification system (ENS) can improve communication times and patient turnaround in the Emergency Department (ED). METHODS: In 01/2019, an ENS allowing direct communication between radiology and ED was installed. Starting in 10/2019, CTPAs were processed by a deep learning (DL)-powered algorithm for detection of PE. CTPAs acquired between 04/2018 and 06/2020 (n = 1808) were analysed. To assess the impact of the ENS and the DL-algorithm, radiology report reading times (RRT), radiology report communication time (RCT), time to anticoagulation (TTA), and patient turnaround times (TAT) in the ED were compared for three consecutive time periods. Performance measures of the algorithm were calculated on a per exam level (sensitivity, specificity, PPV, NPV, F1-score), with written reports and exam review as ground truth. RESULTS: Sensitivity of the algorithm was 79.6 % (95 %CI:70.8-87.2%), specificity 95.0 % (95 %CI:92.0-97.1%), PPV 82.2 % (95 %CI:73.9-88.3), and NPV 94.1 % (95 %CI:91.4-96 %). There was no statistically significant reduction of any of the observed times (RRT, RCT, TTA, TAT). CONCLUSION: DL-assisted detection of PE in CTPAs and ENS-assisted communication of results to referring physicians technically work. However, the mere clinical introduction of these tools, even if they exhibit a good performance, is not sufficient to achieve significant effects on clinical performance measures.

Machine learning automatically detects COVID-19 using chest CTs in a large multicenter cohort.

OBJECTIVES: To investigate machine learning classifiers and interpretable models using chest CT for detection of COVID-19 and differentiation from other pneumonias, interstitial lung disease (ILD) and normal CTs. METHODS: Our retrospective multi-institutional study obtained 2446 chest CTs from 16 institutions (including 1161 COVID-19 patients). Training/validation/testing cohorts included 1011/50/100 COVID-19, 388/16/33 ILD, 189/16/33 other pneumonias, and 559/17/34 normal (no pathologies) CTs. A metric-based approach for the classification of COVID-19 used interpretable features, relying on logistic regression and random forests. A deep learning-based classifier differentiated COVID-19 via 3D features extracted directly from CT attenuation and probability distribution of airspace opacities. RESULTS: Most discriminative features of COVID-19 are the percentage of airspace opacity and peripheral and basal predominant opacities, concordant with the typical characterization of COVID-19 in the literature. Unsupervised hierarchical clustering compares feature distribution across COVID-19 and control cohorts. The metrics-based classifier achieved AUC = 0.83, sensitivity = 0.74, and specificity = 0.79 versus respectively 0.93, 0.90, and 0.83 for the DL-based classifier. Most of ambiguity comes from non-COVID-19 pneumonia with manifestations that overlap with COVID-19, as well as mild COVID-19 cases. Non-COVID-19 classification performance is 91% for ILD, 64% for other pneumonias, and 94% for no pathologies, which demonstrates the robustness of our method against different compositions of control groups. CONCLUSIONS: Our new method accurately discriminates COVID-19 from other types of pneumonia, ILD, and CTs with no pathologies, using quantitative imaging features derived from chest CT, while balancing interpretability of results and classification performance and, therefore, may be useful to facilitate diagnosis of COVID-19. KEY POINTS: • Unsupervised clustering reveals the key tomographic features including percent airspace opacity and peripheral and basal opacities most typical of COVID-19 relative to control groups. • COVID-19-positive CTs were compared with COVID-19-negative chest CTs (including a balanced distribution of non-COVID-19 pneumonia, ILD, and no pathologies). Classification accuracies for COVID-19, pneumonia, ILD, and CT scans with no pathologies are respectively 90%, 64%, 91%, and 94%. • Our deep learning (DL)-based classification method demonstrates an AUC of 0.93 (sensitivity 90%, specificity 83%). Machine learning methods applied to quantitative chest CT metrics can therefore improve diagnostic accuracy in suspected COVID-19, particularly in resource-constrained environments.

Vascular Abnormalities Detected with Chest CT in COVID-19: Spectrum, Association with Parenchymal Lesions, Cardiac Changes, and Correlation with Clinical Severity (COVID-CAVA Study).

Although vascular abnormalities are thought to affect coronavirus disease 2019 (COVID-19) patients' outcomes, they have not been thoroughly characterized in large series of unselected patients. The Swiss national registry coronavirus-associated vascular abnormalities (CAVA) is a multicentric cohort of patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection who underwent a clinically indicated chest computed tomography (CT) aiming to assess the prevalence, severity, distribution, and prognostic value of vascular and non-vascular-related CT findings. Clinical outcomes, stratified as outpatient treatment, inpatient without mechanical ventilation, inpatient with mechanical ventilation, or death, will be correlated with CT and biological markers. The main objective is to assess the prevalence of cardiovascular abnormalities-including pulmonary embolism (PE), cardiac morphology, and vascular congestion. Secondary objectives include the predictive value of cardiovascular abnormalities in terms of disease severity and fatal outcome and the association of lung inflammation with vascular abnormalities at the segmental level. New quantitative approaches derived from CT imaging are developed and evaluated in this study. Patients with and without vascular abnormalities will be compared, which is supposed to provide insights into the prognostic role and potential impact of such signs on treatment strategy. Results are expected to enable the development of an integrative score combining both clinical data and imaging findings to predict outcomes.

Prediction of Patient Management in COVID-19 Using Deep Learning-Based Fully Automated Extraction of Cardiothoracic CT Metrics and Laboratory Findings.

OBJECTIVE: To extract pulmonary and cardiovascular metrics from chest CTs of patients with coronavirus disease 2019 (COVID-19) using a fully automated deep learning-based approach and assess their potential to predict patient management. MATERIALS AND METHODS: All initial chest CTs of patients who tested positive for severe acute respiratory syndrome coronavirus 2 at our emergency department between March 25 and April 25, 2020, were identified (n = 120). Three patient management groups were defined: group 1 (outpatient), group 2 (general ward), and group 3 (intensive care unit [ICU]). Multiple pulmonary and cardiovascular metrics were extracted from the chest CT images using deep learning. Additionally, six laboratory findings indicating inflammation and cellular damage were considered. Differences in CT metrics, laboratory findings, and demographics between the patient management groups were assessed. The potential of these parameters to predict patients' needs for intensive care (yes/no) was analyzed using logistic regression and receiver operating characteristic curves. Internal and external validity were assessed using 109 independent chest CT scans. RESULTS: While demographic parameters alone (sex and age) were not sufficient to predict ICU management status, both CT metrics alone (including both pulmonary and cardiovascular metrics; area under the curve [AUC] = 0.88; 95% confidence interval [CI] = 0.79-0.97) and laboratory findings alone (C-reactive protein, lactate dehydrogenase, white blood cell count, and albumin; AUC = 0.86; 95% CI = 0.77-0.94) were good classifiers. Excellent performance was achieved by a combination of demographic parameters, CT metrics, and laboratory findings (AUC = 0.91; 95% CI = 0.85-0.98). Application of a model that combined both pulmonary CT metrics and demographic parameters on a dataset from another hospital indicated its external validity (AUC = 0.77; 95% CI = 0.66-0.88). CONCLUSION: Chest CT of patients with COVID-19 contains valuable information that can be accessed using automated image analysis. These metrics are useful for the prediction of patient management.

Development and clinical implementation of tailored image analysis tools for COVID-19 in the midst of the pandemic: The synergetic effect of an open, clinically embedded software development platform and machine learning.

PURPOSE: During the emerging COVID-19 pandemic, radiology departments faced a substantial increase in chest CT admissions coupled with the novel demand for quantification of pulmonary opacities. This article describes how our clinic implemented an automated software solution for this purpose into an established software platform in 10 days. The underlying hypothesis was that modern academic centers in radiology are capable of developing and implementing such tools by their own efforts and fast enough to meet the rapidly increasing clinical needs in the wake of a pandemic. METHOD: Deep convolutional neural network algorithms for lung segmentation and opacity quantification on chest CTs were trained using semi-automatically and manually created ground-truth (Ntotal = 172). The performance of the in-house method was compared to an externally developed algorithm on a separate test subset (N = 66). RESULTS: The final algorithm was available at day 10 and achieved human-like performance (Dice coefficient = 0.97). For opacity quantification, a slight underestimation was seen both for the in-house (1.8 %) and for the external algorithm (0.9 %). In contrast to the external reference, the underestimation for the in-house algorithm showed no dependency on total opacity load, making it more suitable for follow-up. CONCLUSIONS: The combination of machine learning and a clinically embedded software development platform enabled time-efficient development, instant deployment, and rapid adoption in clinical routine. The algorithm for fully automated lung segmentation and opacity quantification that we developed in the midst of the COVID-19 pandemic was ready for clinical use within just 10 days and achieved human-level performance even in complex cases.

Automated detection of pulmonary embolism in CT pulmonary angiograms using an AI-powered algorithm.

OBJECTIVES: To evaluate the performance of an AI-powered algorithm for the automatic detection of pulmonary embolism (PE) on chest computed tomography pulmonary angiograms (CTPAs) on a large dataset. METHODS: We retrospectively identified all CTPAs conducted at our institution in 2017 (n = 1499). Exams with clinical questions other than PE were excluded from the analysis (n = 34). The remaining exams were classified into positive (n = 232) and negative (n = 1233) for PE based on the final written reports, which defined the reference standard. The fully anonymized 1-mm series in soft tissue reconstruction served as input for the PE detection prototype algorithm that was based on a deep convolutional neural network comprising a Resnet architecture. It was trained and validated on 28,000 CTPAs acquired at other institutions. The result series were reviewed using a web-based feedback platform. Measures of diagnostic performance were calculated on a per patient and a per finding level. RESULTS: The algorithm correctly identified 215 of 232 exams positive for pulmonary embolism (sensitivity 92.7%; 95% confidence interval [CI] 88.3-95.5%) and 1178 of 1233 exams negative for pulmonary embolism (specificity 95.5%; 95% CI 94.2-96.6%). On a per finding level, 1174 of 1352 findings marked as embolus by the algorithm were true emboli. Most of the false positive findings were due to contrast agent-related flow artifacts, pulmonary veins, and lymph nodes. CONCLUSION: The AI prototype algorithm we tested has a high degree of diagnostic accuracy for the detection of PE on CTPAs. Sensitivity and specificity are balanced, which is a prerequisite for its clinical usefulness. KEY POINTS: • An AI-based prototype algorithm showed a high degree of diagnostic accuracy for the detection of pulmonary embolism on CTPAs. • It can therefore help clinicians to automatically prioritize exams with a high suspection of pulmonary embolism and serve as secondary reading tool. • By complementing traditional ways of worklist prioritization in radiology departments, this can speed up the diagnostic and therapeutic workup of patients with pulmonary embolism and help to avoid false negative calls.

Automated Quantification of CT Patterns Associated with COVID-19 from Chest CT.

PURPOSE: To present a method that automatically segments and quantifies abnormal CT patterns commonly present in coronavirus disease 2019 (COVID-19), namely ground glass opacities and consolidations. MATERIALS AND METHODS: In this retrospective study, the proposed method takes as input a non-contrasted chest CT and segments the lesions, lungs, and lobes in three dimensions, based on a dataset of 9749 chest CT volumes. The method outputs two combined measures of the severity of lung and lobe involvement, quantifying both the extent of COVID-19 abnormalities and presence of high opacities, based on deep learning and deep reinforcement learning. The first measure of (PO, PHO) is global, while the second of (LSS, LHOS) is lobewise. Evaluation of the algorithm is reported on CTs of 200 participants (100 COVID-19 confirmed patients and 100 healthy controls) from institutions from Canada, Europe and the United States collected between 2002-Present (April, 2020). Ground truth is established by manual annotations of lesions, lungs, and lobes. Correlation and regression analyses were performed to compare the prediction to the ground truth. RESULTS: Pearson correlation coefficient between method prediction and ground truth for COVID-19 cases was calculated as 0.92 for PO (P < .001), 0.97 for PHO(P < .001), 0.91 for LSS (P < .001), 0.90 for LHOS (P < .001). 98 of 100 healthy controls had a predicted PO of less than 1%, 2 had between 1-2%. Automated processing time to compute the severity scores was 10 seconds per case compared to 30 minutes required for manual annotations. CONCLUSION: A new method segments regions of CT abnormalities associated with COVID-19 and computes (PO, PHO), as well as (LSS, LHOS) severity scores.

Towards automated generation of curated datasets in radiology: Application of natural language processing to unstructured reports exemplified on CT for pulmonary embolism.

PURPOSE: To design and evaluate a self-trainable natural language processing (NLP)-based procedure to classify unstructured radiology reports. The method enabling the generation of curated datasets is exemplified on CT pulmonary angiogram (CTPA) reports. METHOD: We extracted the impressions of CTPA reports created at our institution from 2016 to 2018 (n = 4397; language: German). The status (pulmonary embolism: yes/no) was manually labelled for all exams. Data from 2016/2017 (n = 2801) served as a ground truth to train three NLP architectures that only require a subset of reference datasets for training to be operative. The three architectures were as follows: a convolutional neural network (CNN), a support vector machine (SVM) and a random forest (RF) classifier. Impressions of 2018 (n = 1377) were kept aside and used for general performance measurements. Furthermore, we investigated the dependence of classification performance on the amount of training data with multiple simulations. RESULTS: The classification performance of all three models was excellent (accuracies: 97 %-99 %; F1 scores 0.88-0.97; AUCs: 0.993-0.997). Highest accuracy was reached by the CNN with 99.1 % (95 % CI 98.5-99.6 %). Training with 470 labelled impressions was sufficient to reach an accuracy of > 93 % with all three NLP architectures. CONCLUSION: Our NLP-based approaches allow for an automated and highly accurate retrospective classification of CTPA reports with manageable effort solely using unstructured impression sections. We demonstrated that this approach is useful for the classification of radiology reports not written in English. Moreover, excellent classification performance is achieved at relatively small training set sizes.

Lethal COVID-19: Radiologic-Pathologic Correlation of the Lungs.

PURPOSE: The purpose of this retrospective study was to correlate CT patterns of fatal cases of coronavirus disease 2019 (COVID-19) with postmortem pathology observations. MATERIALS AND METHODS: The study included 70 lung lobes of 14 patients who died of reverse-transcription polymerase chain reaction-confirmed COVID-19. All patients underwent antemortem CT and autopsy between March 9 and April 30, 2020. Board-certified radiologists and pathologists performed lobewise correlations of pulmonary observations. In a consensus reading, 267 radiologic and 257 histopathologic observations of the lungs were recorded and systematically graded according to severity. These observations were matched and evaluated. RESULTS: Predominant CT observations were ground-glass opacities (GGO) (59/70 lobes examined) and areas of consolidation (33/70). The histopathologic observations were consistent with diffuse alveolar damage (70/70) and capillary dilatation and congestion (70/70), often accompanied by microthrombi (27/70), superimposed acute bronchopneumonia (17/70), and leukocytoclastic vasculitis (7/70). Four patients had pulmonary emboli. Bronchial wall thickening at CT histologically corresponded with acute bronchopneumonia. GGOs and consolidations corresponded with mixed histopathologic observations, including capillary dilatation and congestion, interstitial edema, diffuse alveolar damage, and microthrombosis. Vascular alterations were prominent observations at both CT and histopathology. CONCLUSION: A significant proportion of GGO correlated with the pathologic processes of diffuse alveolar damage, capillary dilatation and congestion, and microthrombosis. Our results confirm the presence and underline the importance of vascular alterations as key pathophysiologic drivers in lethal COVID-19.Supplemental material is available for this article.© RSNA, 2020.

Automated Quantification of CT Patterns Associated with COVID-19 from Chest CT.

PURPOSE: To present a method that automatically segments and quantifies abnormal CT patterns commonly present in coronavirus disease 2019 (COVID-19), namely ground glass opacities and consolidations. MATERIALS AND METHODS: In this retrospective study, the proposed method takes as input a non-contrasted chest CT and segments the lesions, lungs, and lobes in three dimensions, based on a dataset of 9749 chest CT volumes. The method outputs two combined measures of the severity of lung and lobe involvement, quantifying both the extent of COVID-19 abnormalities and presence of high opacities, based on deep learning and deep reinforcement learning. The first measure of (PO, PHO) is global, while the second of (LSS, LHOS) is lobe-wise. Evaluation of the algorithm is reported on CTs of 200 participants (100 COVID-19 confirmed patients and 100 healthy controls) from institutions from Canada, Europe and the United States collected between 2002-Present (April 2020). Ground truth is established by manual annotations of lesions, lungs, and lobes. Correlation and regression analyses were performed to compare the prediction to the ground truth. RESULTS: Pearson correlation coefficient between method prediction and ground truth for COVID-19 cases was calculated as 0.92 for PO (P < .001), 0.97 for PHO (P < .001), 0.91 for LSS (P < .001), 0.90 for LHOS (P < .001). 98 of 100 healthy controls had a predicted PO of less than 1%, 2 had between 1-2%. Automated processing time to compute the severity scores was 10 seconds per case compared to 30 minutes required for manual annotations. CONCLUSION: A new method segments regions of CT abnormalities associated with COVID-19 and computes (PO, PHO), as well as (LSS, LHOS) severity scores.

Centralized expert HRCT Reading in suspected idiopathic pulmonary fibrosis: Experience from an Eurasian teleradiology program.

PURPOSE: To share experience from a large, ongoing expert reading teleradiology program in Europe and Asia aiming at supporting referring centers to interpret high-resolution computed tomography (HRCT) with respect to presence of Usual Interstitial Pneumonia (UIP)-pattern in patients with suspected Idiopathic Pulmonary Fibrosis (IPF). METHOD: We analyzed data from 01/2014 to 05/2019, including HRCTs from 239 medical centers in 12 European and Asian countries that were transmitted to our Picture Archiving and Communication System (PACS) via a secured internet connection. Structured reports were generated in consensus by a radiologist with over 20 years of experience in thoracic imaging and a pulmonologist with specific expertise in interstitial lung disease according to current guidelines on IPF. Reports were sent to referring physicians. We evaluated patient characteristics, technical issues, report turnaround times and frequency of diagnoses. We also conducted a survey to collect feedback from referring physicians. RESULTS: HRCT image data from 703 patients were transmitted (53.5% male). Mean age was 63.7 years (SD:17). In 35.1% of all cases diagnosis was "UIP"/"Typical UIP". The mean report turnaround time was 1.7 days (SD:2.9). Data transmission errors occurred in 7.1%. Overall satisfaction rate among referring physicians was high (8.4 out of 10; SD:3.2). CONCLUSIONS: This Eurasian teleradiology program demonstrates the feasibility of cross-border teleradiology for the provision of state-of-the-art reporting despite heterogeneity of referring medical centers and challenges like data transmission errors and language barriers. We also point out important factors for success like the usage of structured reporting templates.