Machine Learning for Workflow Applications in Screening Mammography: Systematic Review and Meta-Analysis.

Background Advances in computer processing and improvements in data availability have led to the development of machine learning (ML) techniques for mammographic imaging. Purpose To evaluate the reported performance of stand-alone ML applications for screening mammography workflow. Materials and Methods Ovid Embase, Ovid Medline, Cochrane Central Register of Controlled Trials, Scopus, and Web of Science literature databases were searched for relevant studies published from January 2012 to September 2020. The study was registered with the PROSPERO International Prospective Register of Systematic Reviews (protocol no. CRD42019156016). Stand-alone technology was defined as a ML algorithm that can be used independently of a human reader. Studies were quality assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 and the Prediction Model Risk of Bias Assessment Tool, and reporting was evaluated using the Checklist for Artificial Intelligence in Medical Imaging. A primary meta-analysis included the top-performing algorithm and corresponding reader performance from which pooled summary estimates for the area under the receiver operating characteristic curve (AUC) were calculated using a bivariate model. Results Fourteen articles were included, which detailed 15 studies for stand-alone detection (n = 8) and triage (n = 7). Triage studies reported that 17%-91% of normal mammograms identified could be read by adapted screening, while "missing" an estimated 0%-7% of cancers. In total, an estimated 185 252 cases from three countries with more than 39 readers were included in the primary meta-analysis. The pooled sensitivity, specificity, and AUC was 75.4% (95% CI: 65.6, 83.2; P = .11), 90.6% (95% CI: 82.9, 95.0; P = .40), and 0.89 (95% CI: 0.84, 0.98), respectively, for algorithms, and 73.0% (95% CI: 60.7, 82.6), 88.6% (95% CI: 72.4, 95.8), and 0.85 (95% CI: 0.78, 0.97), respectively, for readers. Conclusion Machine learning (ML) algorithms that demonstrate a stand-alone application in mammographic screening workflows achieve or even exceed human reader detection performance and improve efficiency. However, this evidence is from a small number of retrospective studies. Therefore, further rigorous independent external prospective testing of ML algorithms to assess performance at preassigned thresholds is required to support these claims. ©RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Whitman and Moseley in this issue.

Abnormal Pulmonary Venous Filling: An Adjunct Feature in the Computed Tomography Pulmonary Angiogram Assessment of Chronic Thromboembolic Pulmonary Hypertension.

Background Hypodense filling defects within the pulmonary veins on computed tomography described as pulmonary vein sign (PVS) have been noted in acute pulmonary embolism and shown to be associated with poor prognosis. We evaluated venous flow abnormalities in chronic thromboembolic pulmonary hypertension (CTEPH) to determine its usefulness in the computed tomography assessment of CTEPH. Methods and Results Blinded retrospective computed tomography analysis of 50 proximal CTEPH cases and 3 control groups-50 acute pulmonary embolism, 50 nonthromboembolic cohort, and 50 pulmonary arterial hypertension. Venous flow reduction was assessed by the following: (1) presence of a filling defect of at least 2 cm in a pulmonary vein draining into the left atrium, and (2) left atrium attenuation (>160 Hounsfield units). PVS was most prevalent in CTEPH. Compared with all controls, sensitivity and specificity of PVS for CTEPH is 78.0% and 85.3% (95% CI, 64.0-88.5 and 78.6-90.6, respectively) versus 34.0% and 70.7% (95% CI, 21.2-48.8 and 62.7-77.8) in acute pulmonary embolism, 8.0% and 62% (95% CI, 2.2-19.2 and 53.7-69.8) in nonthromboembolic and 2.0% and 60% (95% CI, 0.1-10.7 and 51.7-67.9) in pulmonary arterial hypertension. In CTEPH, lobar and segmental arterial occlusive disease was most commonly associated with corresponding absent venous flow. PVS detection was highly reproducible (Kappa=0.96, 95% CI, 0.90-1.01, P<0.001). Conclusions PVS is easy to detect with higher sensitivity and specificity in CTEPH compared with acute pulmonary embolism and is not a feature of pulmonary arterial hypertension. Asymmetric enhancement of pulmonary veins may serve as an additional parameter in the computed tomography assessment of CTEPH and can be used to differentiate CTEPH from pulmonary arterial hypertension.

A zero coronary artery calcium score in patients with stable chest pain is associated with a good prognosis, despite risk of non-calcified plaques.

Objectives: To estimate the prevalence of non-calcified coronary artery disease (CAD) in patients with suspected stable angina and a zero coronary artery calcification (CAC) score, and to assess the prognostic significance of a zero CAC in these symptomatic patients. Methods: In this prospective cohort study, consecutive patients with stable chest pain underwent CAC scoring ± CT coronary angiography (CTCA) as part of routine clinical care at a single tertiary centre over 7 years. Major adverse cardiac event (MACE) was defined as cardiac death, non-fatal myocardial infarction and/or non-elective revascularisation. Results: A total of 915 of 1753 (52.2%) patients (mean age 56.8 ± 12.0 years; 46.2% male) had a zero CAC score. Of the 751 (82.1%) patients with a zero CAC in whom CTCA was performed, 674 (89.7%) had normal coronary arteries, 63 (8.4%) had non-calcified CAD with < 50% stenosis and 14 (1.9%) had ≥ 50% stenosis in at least one coronary artery. The negative predictive value of a zero CAC for excluding a ≥ 50% CTCA stenosis was 98.1%. Over a median follow-up period of 2.2 years (range 1.0-7.0 years), the absolute annualised rates of MACE were as follows: zero CAC 1.9 per 1000 person-years and non-zero CAC 7.4 per 1000 person-years (HR 3.8, p = 0.009). However, after adjusting for age, gender and cardiovascular risk factors using a multivariable Cox proportional hazards model, there was no statistically significant difference in the risk of MACE between the two patient cohorts (p = 0.19). After adjusting for age, gender and cardiovascular risk factors, the HR for all-cause mortality among the zero CAC cohort vers non-zero CAC was 2.1 (p = 0.27). Conclusion: A zero CAC score in patients undergoing CT scanning for suspected stable angina has a high negative predictive value for the exclusion of obstructive CAD and is associated with a good medium-term prognosis.