Thoracic
Using Quantitative Computed Tomographic Imaging to Understand Chronic Obstructive Pulmonary Disease and Fibrotic Interstitial Lung Disease. State of the Art and Future Directions.
Castillo-Saldana, D., et al. (2020). “Using Quantitative Computed Tomographic Imaging to Understand Chronic Obstructive Pulmonary Disease and Fibrotic Interstitial Lung Disease: State of the Art and Future Directions.” J Thorac Imaging 35(4): 246-254.
Key Words: COPD, ILD, Quantitative CT
Researchers from St. Paul’s Hospital and the University of British Columbia report interesting applications for Quantitative CT (QCT) in the realm of chronic obstructive lung disease (COPD) and interstitial lung disease (ILD). Ryerson et al pose that QCT applications can be used in adjunct to typical Qualitative assessments of COPD and ILD to diagnose and track progression of these two complex disease patterns. QCT can be used in COPD to accurately map areas of emphysema, identify airway thickening and create 3-D reconstructions of the bronchial tree to identify early disease at the terminal bronchioles. Major limitations in the clinical application of QCT in COPD are the exposure to ionizing radiation, CT scanner and patient variability, and significant quality control due to manual post-processing. QCT can play an important role in tracking the progression of ILD by examining discrete pulmonary parenchymal densities and plotting these findings on a density histogram. This objective data could be used in conjunction with qualitative evaluations to provide clinicians with a more precise way to track disease progression and treatment response in a highly morbid patient population. Significant improvements are still needed in these quantitative algorithms before they find surefire clinical applications, but these are exciting developments nonetheless.
COVID-19
Pulmonary Thromboembolism in COVID-19: Venous Thromboembolism or Arterial Thrombosis?
Ferrari, F., et al. (2020). “Pulmonary Thromboembolism in COVID-19: Venous Thromboembolism or Arterial Thrombosis?” RSNA
Cavagna, E., et al. (2020). “Pulmonary Thromboembolism in COVID-19: Venous Thromboembolism or Arterial Thrombosis?” Radiology: Cardiothoracic Imaging 2(4): e200289.
https://pubs.rsna.org/doi/10.1148/ryct.2020200289
Key Words: COVID-19, Pulmonary Embolus, Thromboembolism
Italian radiologists Cavagna et al. provide a window into the association of the novel coronavirus and thromboembolism in severely ill patients. COVID-19 patients are noted to have a higher rate of pulmonary arterial emboli (PE); which is especially concerning, given the higher frequency of critically ill patients at baseline. In this retrospective study, 109 patients with COVID-19 who exhibited symptoms concerning for PE and had a CT pulmonary angiogram (CTPA) to rule out pulmonary embolus were enrolled. Patients were assigned a semiquantitative score (0-5) to estimate pulmonary involvement of the 5 lobes of the lung. Patients with higher CT lesion score and abnormal lab values such as LDH, CRP and d-dimer were found to have higher rates of pulmonary thromboembolism. These hypercoagulable phenomena preferentially involved the segmental (90.2%) and sub-segmental arteries (61%) of the pulmonary parenchyma most affected by COVID. The authors postulate that these are pulmonary artery thromboses secondary to local lung inflammation and hypercoagulable state rather than emboli. Although the study lacks the power to definitively verify this statement, the authors draw upon evidence from other studies that further justify their radical claim. Hopefully future studies with more data points will be available to validate their hypothesis.
Radiological Society of North America Chest CT Classification System for Reporting COVID-19 Pneumonia: Interobserver Variability and Correlation with RT-PCR
Jaegere, T. M. H. d., et al. (2020). “Radiological Society of North America Chest CT Classification System for Reporting COVID-19 Pneumonia: Interobserver Variability and Correlation with RT-PCR.” Radiology: Cardiothoracic Imaging 2(3): e200213.
Keywords: COVID-19, RSNA Classification, CORADS
https://pubs.rsna.org/doi/10.1148/ryct.2020200213
Radiologists from the Zuyderland Medical Center in the Netherlands investigate interobserver agreement within Radiologists while using the RSNA COVID-19 Chest CT classification system and COVID-19 Reporting and Data System (CO-RADS); mapping these results with patients’ PCR results. In a retrospective review of 96 patient’s chest CT’s, 2 attending Radiologist and one 5th year resident utilized the RSNA Chest CT Classification system and the CORADS classification system to test interobserver variability. Radiologists were tasked with classifying CT findings as “typical”, “indeterminate”, “atypical” or “negative” for the presence of COVID. They found substantial interobserver agreement between the attendings and moderate agreement between the attendings’ and resident’s reads. This data was then compared to patients’ RT-PCR COVID test results. The 45 patients who tested positive were stratified by CT findings and the positivity rate is reported as “Typical” 76.9-96.6%, “Indeterminate” 51.2-64.1%, “Atypical 2.8-5.3% and “Negative” 20-25%. A low positive rate in the “atypical” group is explained by the presence of findings associated with other lung diseases. The curiously high rate of positivity in the “negative” group is concerning. The authors pose that this group may have patients who have yet to manifest signs associated with COVID-19 infection, as imaging findings can change over time. The authors also utilized the CO-RADs classification system and achieved similar interobserver agreement with non-negligible positives in the CO-RADS 1 group. These results likely mimic clinical practice, as symptomatic patients will present at varying times in the disease process, such that CT seems to have a limited role in the screening of Coronavirus. As severe COVID-19 infections are associated with significant medical morbidity and mortality, CT still plays a valuable role in patient management.
Cardiac
Feasibility of Cardiovascular Four-dimensional Flow MRI during Exercise in Healthy Participants.
MacDonald, J.A., et al. (2020). “Feasibility of Cardiovascular Four-dimensional Flow MRI during Exercise in Healthy Participants.” Radiology: Cardiothoracic Imaging 2(3):e190033
Keywords: Cardiac MRI, 4D-Flow
https://pubs.rsna.org/doi/10.1148/ryct.2020190033
Macdonald et al. examine the effectiveness of free-breathing, 4-D flow MRI during exercise. In a healthy cohort of 10 individuals cardiac MRI was performed at rest and during stress, in which a standardized exercise protocol was implemented on an MRI compatible stair stepper device while supine in the bore of the magnet. These studies were retrospectively gated to account for breath motion artifact and EKG-gated to account for cardiac motion. This group used a 4-D flow protocol to assess stroke volume (SV) and ventricular kinetic energy (KE) in systole and diastole during rest and stress examinations. They found that cardiac output predictably increased during exercise when compared to rest. Interestingly though, they found that right ventricular kinetic energy (KERV) was significantly increased from rest, while left ventricular kinetic energy (KELV) and (SV) were not. SV has previously been documented to be unaffected during supine exercise. KERV was noted to be statistically significantly higher in systole compared to KELV, and not different during diastole. As the left ventricular KE is not significantly affected, 4-D stress MRI may have limited applications. However, this may become a valuable tool for the evaluation of right heart disease as the KERV is affected. There are notable limitations for applications as current practices stand secondary to poor interobserver and intraobserver reproducibility in relation to ventricular masking and KE estimation.
Machine Learning
Value of Machine Learning-based Coronary CT Fractional Flow Reserve Applied to Triple-Rule-Out CT Angiography in Acute Chest Pain.
Schoepf, U.J., et al. (2020) “Value of Machine Learning-based Coronary CT Fractional Flow Reserve Applied to Triple-Rule-Out CT Angiography in Acute Chest Pain.” Radiology: Cardiothoracic Imaging 2(3):e190137
Keywords: Machine Learning, Triple Rule Out (TRO), Fractional Flow Reserve (FFR), CT Coronary Angiography
https://pubs.rsna.org/doi/10.1148/ryct.2020190137
Radiologists at the Medical University of South Carolina describe an application for machine learning in estimating fractional flow reserve during triple rule out CT angiography. In a retrospective review encompassing 159 patients, Schoepf et al. reviewed 159 Triple-Rule-Out CT Angiograms (TRO) for patients who had low to moderate probability of cardiac disease. The fractional flow reserve (FFR) was then calculated using a deep learning algorithm. FFR exhibits high diagnostic accuracy in evaluating significant obstructive coronary disease, and has been used in the ambulatory setting to assess coronary artery disease (CAD). This group sought to use non-commercially available on-site FFR calculation the emergency setting to identify patients with hemodynamically significant stenoses and used coronary intervention or major adverse cardiovascular events (MACE) as endpoints. An FFR <0.8 was considered hemodynamically significant stenosis. Of the patients who received TRO, 55% had significant stenosis (>50%) and 5% of had severe stenosis (>70%). 52% of the study subjects exhibited congruent findings in FFR and TRO; of these, 55% were found to have coronary significant coronary stenosis and an FFR <0.8, while 45% had FFR>0.8 and no significant stenosis on TRO. In chart review it was noted that coronary revascularization or significant MACE event occurred in 27.5% of patients imaged, and that those with an FFR <0.8 were significantly more likely to meet one of these two endpoints than their peers. It is noted that 67% of patients who received a TRO underwent further downstream testing, a relatively common phenomenon. The authors pose that an FFR <0.8 may be useful to delineate patients who are at increased risk and would warrant further testing. It is noted that 4 patients had FFR >0.8 and underwent coronary revascularization, so these 4 patients may have missed out on therapeutic intervention had FFR been used as a cutoff value for further referral. It is also noted that in 54% of patients who underwent both TRO and SPECT and 48% of patients who underwent TRO and stress echocardiography, there were discordant findings; 3 of these patients had an FFR >0.8 and a positive result on stress SPECT/CT. As this study was retrospective and no treatment decisions were made using the FFR <0.8 value as a reference point, it is unclear how many patients would ultimately have a negative outcome. It seems that adding FFR to TRO in patients with low to moderate-risk CAD may be a valuable addition to the assessment of acute chest pain.
CT-based True- and False-Lumen Segmentation in Type B Aortic Dissection Using Machine Learning.
Hahn, L.D., et al. “CT-based True- and False-Lumen Segmentation in Type B Aortic Dissection Using Machine Learning.” Radiology: Cardiothoracic Imaging 2020; 2(3): e190179
Keywords: Aortic Dissection, Machine Learning
https://pubs.rsna.org/doi/10.1148/ryct.2020190179
Researchers at Stanford utilized a stepwise convoluted neural network (CNN) to automatically generate multiplanar reconstructions (MPRs) of uncomplicated Type B aortic dissection and then to segment the true (TL) and false lumens (FL). Hahn, L.D. et al. retrospectively identified 45 patients with a cumulative 153 CT angiograms of the aorta (CTA). Patients were randomly assigned to a training set, a validation set and a test set of studies, which were then fed into the CNN’s stepwise segmentation. The steps for segmentation included identifying the aorta from background in the axial plane, determining the aortic centerline by utilizing the average density within the aorta, generating MPR’s in an orthogonal plane from the centerline, then segmenting the aortic dissection into TL and FL, and finally converting the MPR’s back to the axial plane. The automatic segmentation was compared to internal controls (the test set) and to manual aortic segmentation by expert Cardiothoracic Radiologists. End points for this study included the dice similarity coefficient (DSC) and the average Euclidean distance between edges in mm. Researchers found that there was no significant difference between the validation set and the test set of data with the average DSC ranging from 0.873-0.9, suggesting highly accurate segmentation. The average Euclidean distance was <3mm. Automated segmentation also generated diameter plots for the TL, FL and aorta throughout its length giving Radiologists the ability to quantitively assess volumes. This is highly applicable to patient care, in that aortic diameter is used to triage aortic intervention. Current clinical applications are still somewhat far off given the need for manual input by an expert user, as errors can occur in the identification of the TL and FL (reported on 3.8% of studies). The authors also propose that the CNN may have been introduced bias since the same study population was used in testing and validation. As data sets and patient population grows the CNN should be able to differentiate a wider phenotypic variety of dissection. Most AI algorithms also tend to struggle in real-life scenarios due to the innate variability in CT scanners, patient body habitus/anatomy and noise. As AI continues to improve, this will undoubtedly become a reliable tool for clinicians to monitor uncomplicated type B aortic dissections.
References
