Automatic prediction of left cardiac chamber enlargement from chest radiographs using convolutional neural network.

OBJECTIVE: To develop deep learning-based cardiac chamber enlargement-detection algorithms for left atrial (DLCE-LAE) and ventricular enlargement (DLCE-LVE), on chest radiographs METHODS: For training and internal validation of DLCE-LAE and -LVE, 5,045 chest radiographs (CRs; 2,463 normal and 2,393 LAE) and 1,012 CRs (456 normal and 456 LVE) matched with the same-day echocardiography were collected, respectively. External validation was performed using 107 temporally independent CRs. Reader performance test was conducted using the external validation dataset by five cardiothoracic radiologists without and with the results of DLCE. Classification performance of DLCE was evaluated and compared with those of the readers and conventional radiographic features, including cardiothoracic ratio, carinal angle, and double contour. In addition, DLCE-LAE was tested on 5,277 CRs from a healthcare screening program cohort. RESULTS: DLCE-LAE showed areas under the receiver operating characteristics curve (AUROCs) of 0.858 on external validation. On reader performance test, DLCE-LAE showed better results than pooled radiologists (AUROC 0.858 vs. 0.651; p < .001) and significantly increased their performance when used as a second reader (AUROC 0.651 vs. 0.722; p < .001). DLCE-LAE also showed a significantly higher AUROC than conventional radiographic findings (AUROC 0.858 vs. 0.535-0.706; all ps < .01). In the healthcare screening cohort, DLCE-LAE successfully detected 71.0% (142/200) CRs with moderate-to-severe LAE (93.5% [29/31] of severe cases), while yielding 11.8% (492/4,184) false-positive rate. DLCE-LVE showed AUROCs of 0.966 and 0.594 on internal and external validation, respectively. CONCLUSION: DLCE-LAE outperformed and improved cardiothoracic radiologists' performance in detecting LAE and showed promise in screening individuals with moderate-to-severe LAE in a healthcare screening cohort. KEY POINTS: • Our deep learning algorithm outperformed cardiothoracic radiologists in detecting left atrial enlargement on chest radiographs. • Cardiothoracic radiologists improved their performance in detecting left atrial enlargement when aided by the algorithm. • On a healthcare-screening cohort, our algorithm detected 71.0% (142/200) radiographs with moderate-to-severe left atrial enlargement while yielding 11.8% (492/4,184) false-positive rate.

Performance of a Deep Learning Algorithm Compared with Radiologic Interpretation for Lung Cancer Detection on Chest Radiographs in a Health Screening Population.

Background The performance of a deep learning algorithm for lung cancer detection on chest radiographs in a health screening population is unknown. Purpose To validate a commercially available deep learning algorithm for lung cancer detection on chest radiographs in a health screening population. Materials and Methods Out-of-sample testing of a deep learning algorithm was retrospectively performed using chest radiographs from individuals undergoing a comprehensive medical check-up between July 2008 and December 2008 (validation test). To evaluate the algorithm performance for visible lung cancer detection, the area under the receiver operating characteristic curve (AUC) and diagnostic measures, including sensitivity and false-positive rate (FPR), were calculated. The algorithm performance was compared with that of radiologists using the McNemar test and the Moskowitz method. Additionally, the deep learning algorithm was applied to a screening cohort undergoing chest radiography between January 2008 and December 2012, and its performances were calculated. Results In a validation test comprising 10 285 radiographs from 10 202 individuals (mean age, 54 years ± 11 [standard deviation]; 5857 men) with 10 radiographs of visible lung cancers, the algorithm's AUC was 0.99 (95% confidence interval: 0.97, 1), and it showed comparable sensitivity (90% [nine of 10 radiographs]) to that of the radiologists (60% [six of 10 radiographs]; P = .25) with a higher FPR (3.1% [319 of 10 275 radiographs] vs 0.3% [26 of 10 275 radiographs]; P < .001). In the screening cohort of 100 525 chest radiographs from 50 070 individuals (mean age, 53 years ± 11; 28 090 men) with 47 radiographs of visible lung cancers, the algorithm's AUC was 0.97 (95% confidence interval: 0.95, 0.99), and its sensitivity and FPR were 83% (39 of 47 radiographs) and 3% (2999 of 100 478 radiographs), respectively. Conclusion A deep learning algorithm detected lung cancers on chest radiographs with a performance comparable to that of radiologists, which will be helpful for radiologists in healthy populations with a low prevalence of lung cancer. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Armato in this issue.

MRI Findings of Spontaneous Intracranial Hypotension: Usefulness of Straight Sinus Distention.

OBJECTIVE. Spontaneous intracranial hypotension (SIH) shows various characteristic MRI findings. We evaluated the usefulness of straight sinus distention compared with transverse sinus distention and also evaluated other MRI findings of SIH. MATERIALS AND METHODS. Forty-three consecutive patients (28 female and 15 male patients) treated for SIH and 43 age- and sex-matched control subjects at two institutions from 2012 through 2014 were included in this study. Two reviewers determined whether the transverse sinus distention sign and straight sinus distention sign were present on MRI. Diagnostic performance values and interobserver agreement were calculated. Reviewers also assessed MRI examinations in consensus for the presence of the following findings: pachymeningeal enhancement, subdural effusion or hematoma, enlargement of the pituitary gland, and downward displacement of the brainstem and tonsils. RESULTS. The sensitivity, specificity, and diagnostic accuracy of the transverse sinus distention sign for SIH were 76.7%, 83.7%, and 80.2%, whereas those of the straight sinus distention sign were 79.1%, 95.4%, and 87.2%, respectively. The specificity of the straight sinus distention sign for SIH was significantly higher (p = 0.025) than that of the transverse sinus distention sign. In addition, the straight sinus distention sign showed substantial agreement (κ = 0.79), whereas the transverse sinus distention sign showed moderate agreement (κ = 0.60). The diagnostic accuracy of the presence of either transverse or straight sinus distention (83.7%) was significantly higher than that of pachymeningeal enhancement (80.2%, p = 0.032). CONCLUSION. The straight sinus distention sign could be helpful for the diagnosis of SIH because it has sensitivity comparable to other imaging findings and higher specificity and higher level of interobserver agreement than other imaging findings.

Off-site evaluation of three-dimensional ultrasound for the diagnosis of thyroid nodules: comparison with two-dimensional ultrasound.

OBJECTIVES: We compared the diagnostic performance of off-site evaluation between prospectively obtained 3D and 2D ultrasound for thyroid nodules. METHODS: 3D and 2D ultrasonographies were preoperatively obtained from 85 consecutive patients (mean age, 51 years; age range, 28-83 years) who were referred for a total thyroidectomy. Three radiologists independently evaluated 3D and 2D images of 91 pathologically confirmed thyroid nodules (30 benign and 61 malignant nodules) for nodule characterization. Diagnostic performance, interobserver agreement and time for scanning were compared between 3D and 2D. RESULTS: 3D had significantly higher sensitivities than 2D for predicting malignancy (78.7 % vs. 61.2 %, P < 0.01) and extrathyroidal extension (66.7 % vs. 46.4 %, P = 0.03) in malignancy. In terms of specificities, there were no statistically significant differences between 2D and 3D for predicting malignancy (78.4 % vs. 74.8 %, P = 1.00) and extrathyroidal extension (63.6 % vs. 57.6 %, P = 0.46). With respect to interobserver agreement, 3D showed moderate agreement (κ = 0.53) for predicting extrathyroidal extension in malignancy compared with 2D ultrasound, which showed fair agreement (κ = 0.37). 3D saved time (30 ± 56.52 s) for scanning compared with 2D. CONCLUSION: For off-site evaluation, 3D US is more useful for diagnosis of thyroid nodules than 2D US. KEY POINTS: • 3D had higher sensitivity than 2D for predicting malignancy and extrathyroidal extension. • 3D showed better agreement for predicting extrathyroidal extension in malignancy than 2D. • 3D thyroid ultrasound saved time for scanning compared with 2D. • For off-site evaluation of thyroid nodules, 3D is more useful than 2D.

Usefulness of MR elastography for predicting esophageal varices in cirrhotic patients.

PURPOSE: To evaluate the usefulness of magnetic resonance elastography (MRE) as a noninvasive tool for predicting esophageal varices and identifying high-risk varices. MATERIALS AND METHODS: In all, 126 patients with liver cirrhosis, and who underwent both MRI including MRE of the liver as well as upper gastrointestinal endoscopy for variceal screening within 1 month before or after the MRI, were included in this study. The relationship between the liver stiffness values measured by MRE and the degree of esophageal varices was assessed using Spearman's correlation analysis. In addition, the diagnostic performance of MRE for predicting the presence of varices or high-risk varices (grade≥II) was evaluated using the receiver-operating characteristics (ROC) curves. RESULTS: The mean stiffness values of liver parenchyma measured on MRE were well correlated with the grade of esophageal varices (r=0.63). In addition, the MRE-based liver stiffness values were significantly lower in the lower-risk group than in the higher-risk group (P<0.0001). The area under the ROC curve values of MRE for predicting the presence of varices or high-risk varices (grade≥II) were 0.859 and 0.810, respectively. Using a liver stiffness cutoff value of 5.803 kPa, the sensitivity, specificity, positive predictive value, and negative predictive value for predicting high-grade (≥II) esophageal varices were 96%, 60%, 36%, and 98%, respectively. CONCLUSION: The MRE-based liver stiffness value may be useful for noninvasively predicting esophageal varices and identifying high-risk varices in cirrhotic patients.

Spine MR imaging features of subacute combined degeneration patients.

BACKGROUND AND PURPOSE: Subacute combined degeneration (SCD) is a potentially reversible neurological complication of a vitamin B12 deficiency; therefore, timely diagnosis and appropriate treatment are of great importance. The study was to evaluate the spine MR imaging features of SCD in a series of patients. MATERIALS AND METHODS: Eight patients diagnosed with SCD from 2008 to 2010 comprised the study population. Spine MRIs were available for all eight patients, and three of them had follow-up MRIs after vitamin B12 treatment. Two radiologists evaluated the prevalence of signal intensity abnormality of spinal cord and analyzed the distribution and pattern of the signal change in consensus. And they also evaluated post-treatment MRI to find interval change. RESULTS: Seven of eight patients showed abnormal hyperintensity within posterior aspect of spinal cord on T2-weighted images. The spinal cord abnormalities were seen at cervical spine in five patients (62.5 %) and at thoracic spine in the other two patients (25 %). For patients with cervical spinal cord abnormalities, axial T2-weighted images showed symmetric linear T2-hyperintensity as an "inverted V" at cervical spinal cord. For patients with thoracic spinal cord abnormalities, the abnormal signal intensity looked bilateral paired nodular T2-hyperintensity as "dumbbell" or "binoculars" at thoracic spinal cord. Follow-up MRIs after vitamin B12 treatment showed interval resolution of the areas of abnormal T2-hyperintensity in all. CONCLUSION: Symmetric T2-hyperintensity within dorsal column of spinal cord is commonly seen in SCD patients with a linear pattern in the cervical spine and a nodular pattern in the thoracic spine.

Gadoxetic acid-enhanced MRI with MR cholangiography for the preoperative evaluation of bile duct cancer.

PURPOSE: To assess the diagnostic accuracy of gadoxetic acid-enhanced biliary MRI with MR cholangiography (MRC) in the preoperative evaluation of bile duct cancer (BDC) staging and resectability. MATERIALS AND METHODS: Seventy-three patients with BDC who underwent gadoxetic acid-enhanced biliary MRI and MRC, were included in this study. Two abdominal radiologists evaluated the biliary MRI findings regarding the tumor extent, vascular involvement, lymph node metastasis, and tumor resectability. The results were compared with the surgical and pathology findings which were used as the standard reference. The diagnostic performance of the MRI was evaluated using receiver operating characteristics (ROC) analysis. In addition, to determine whether the hepatobiliary phase images had been successfully obtained, the enhancement percentage of the hepatic parenchyma was measured on the portal venous images (PVI) and hepatobiliary phase images (HBPI), respectively. RESULTS: The overall accuracy of the two reviewers for determining the tumor resectability was 61.6% and 83.5%, respectively. The Az values were 0.802 for reviewer 1 and 0.892 for reviewer 2 in the evaluation of the secondary biliary confluence tumor involvement and 0.773 for reviewer 1 and 0.846 for reviewer 2 in the evaluation of the intrapancreatic bile duct involvement. In the evaluation of the vascular involvement, the Az values were 0.718 and 0.906, respectively, for the hepatic artery evaluation and 0.55 and 0.88, respectively, for the portal vein evaluation. For assessment of lymph node metastasis, the overall accuracy was 69.6% and 79.7%, respectively. The mean enhancement percentages of hepatic parenchyma on PVI and HBPI were 39.3% and 65.9%, respectively (P % 0.05), and 49 of 73 patients (67.1%) showed higher enhancement percentage on HBPI than on PVI CONCLUSION: Gadoxetic acid-enhanced MRI with MRC is a reliable diagnostic method for assessing the tumor extent and resectability of BDC.

Percutaneous intervention of the C2 vertebral body using a CT-guided posterolateral approach.

OBJECTIVE: The objective of this article is to present two cases of percutaneous biopsy and one case of vertebroplasty of the C2 vertebral body using a CT-guided posterolateral approach under local anesthesia. CONCLUSION: The CT-guided posterolateral approach was safe, feasible, and effective for percutaneous intervention of the C2 vertebral body.

Hyperintense Thyroid Incidentaloma on Time of Flight Magnetic Resonance Angiography.

Background: The purpose of this study was to evaluate the clinical significance of thyroid incidentaloma with hypersignal intensity on the time of flight magnetic resonance (TOF-MR) angiography and correlation with ultrasound (US). Methods: We retrospectively reviewed 3,505 non-contrast TOF-MR angiography performed at our institution between September 2014 and May 2017. Two radiologists correlated the thyroid incidentalomas detected on TOF-MR angiography with US features that were obtained within a three-month interval between MR and US examinations in consensus. Results: The prevalence of hyperintense thyroid nodules incidentally detected by TOF-MR angiography was 1.2% (43/3,505 patients). Among these, 35 people (77.8%) underwent US examinations, and a total of 45 hyperintense thyroid nodules were detected by US studies. Of these 45 nodules, more than 70% were categorized as benign on US exams. Fine needle aspiration was performed on nine nodules according to indications recommended by the Korean Society of Thyroid Radiology. All except one high-suspicion thyroid nodule were confirmed as benign (Bethesda 2) on cytologic examination. The high-suspicion nodule on US showed a nondiagnostic result (Bethesda 1). However, this nodule collapsed after aspiration of thick colloid. Conclusions: Our study demonstrated that the most hyperintense thyroid nodules detected on TOF-MR angiography were benign. Therefore, if a hyperintense incidentaloma is found on TOF-MR angiography, the thyroid nodule is more likely to be benign. We believe that these findings could offer additional information for further clinical management.

MRI Inter-Reader and Intra-Reader Reliabilities for Assessing Injury Morphology and Posterior Ligamentous Complex Integrity of the Spine According to the Thoracolumbar Injury Classification System and Severity Score.

OBJECTIVE: To evaluate spine magnetic resonance imaging (MRI) inter-reader and intra-reader reliabilities using the thoracolumbar injury classification system and severity score (TLICS) and to analyze the effects of reader experience on reliability and the possible reasons for discordant interpretations. MATERIALS AND METHODS: Six radiologists (two senior, two junior radiologists, and two residents) independently scored 100 MRI examinations of thoracolumbar spine injuries to assess injury morphology and posterior ligamentous complex (PLC) integrity according to the TLICS. Inter-reader and intra-reader agreements were determined and analyzed according to the number of years of radiologist experience. RESULTS: Inter-reader agreement between the six readers was moderate (k = 0.538 for the first and 0.537 for the second review) for injury morphology and fair to moderate (k = 0.440 for the first and 0.389 for the second review) for PLC integrity. No significant difference in inter-reader agreement was observed according to the number of years of radiologist experience. Intra-reader agreements showed a wide range (k = 0.538-0.822 for injury morphology and 0.423-0.616 for PLC integrity). Agreement was achieved in 44 for the first and 45 for the second review about injury morphology, as well as in 41 for the first and 38 for the second review of PLC integrity. A positive correlation was detected between injury morphology score and PLC integrity. CONCLUSION: The reliability of MRI for assessing thoracolumbar spinal injuries according to the TLICS was moderate for injury morphology and fair to moderate for PLC integrity, which may not be influenced by radiologist' experience.

CT imaging spectrum of pancreatic serous tumors: based on new pathologic classification.

PURPOSE: The aim of this study is to retrospectively analyze the variety of CT findings based on new pathologic classification. MATERIALS AND METHODS: During a 10-year period, 59 histopathologically proven pancreatic SCTs and 13 SCTs confirmed with typical image findings and strict clinical criteria were enlisted. Two radiologists analyzed CT images for the following items in consensus: location, size, outer margin, tumor shape, the presence of mural nodule, communication with main pancreatic duct (MPD), the presence and extent of MPD dilatation, calcification, central scar, and attenuation on pre- and post-contrast CT images. In addition, typicality of CT findings was determined. A typical finding was defined as a honeycomb appearance without or with oligocystic portion. In the cases with atypical features, the type of atypical features and differential diagnosis were recorded. For the shape of the tumor, tumors were categorized into the following groups: honeycomb without or with oligocystic, pleomorphic, purely oligolocular, unilocular cystic, hypovascular solid, hypervascular solid without or with oligocystic portion, and fingerlike cystic patterns. RESULTS: 28 SCTs (38.9%) presented a honeycomb appearance with (n=14) or without oligocystic portion (n=14) and were classified as typical cases. The remaining 44 atypical cases (61.1%) presented the following: purely oligolocular pattern in 18; hypervascular solid without (n=7) or with oligocystic portion (n=2) in 9; pleomorphic in 8; unilocular cystic in 7; and fingerlike cystic pattern in 2. Most of the lesions manifesting as hypervascular solid lesions were confused with true solid hypervascular tumors such as neuroendocrine tumors or solid pseudopapillary tumors. For most of the remaining atypical lesions, mucinous cystic neoplasm or branch duct type IPMN were included as a differential diagnosis. CONCLUSION: Serous cystic tumors of the pancreas can have variable CT appearances ranging from compactly solid hypervascular to clearly unilocular cystic, which reflect a varied macroscopic morphology. Only 38.9% of the SCTs show traditional typical CT features whereas a considerable proportion (61.1%) of SCTs showed atypical appearances, providing diagnostic challenges to radiologists.

Gadoxetic acid-enhanced magnetic resonance imaging for differentiating small hepatocellular carcinomas (< or =2 cm in diameter) from arterial enhancing pseudolesions: special emphasis on hepatobiliary phase imaging.

OBJECTIVES: To determine the characteristic enhancing features of hepatocellular carcinoma (HCC) and arterial-enhancing pseudolesion (AEP) on gadoxetic acid (Primovist)-enhanced magnetic resonance imaging (MRI) and to assess its performance compared with that of multirow detector computed tomography (MDCT) for differentiating small HCC (< or =2 cm in diameter) from AEP in cirrhotic liver. MATERIALS AND METHODS: A total of 69 patients with 97 small, arterial enhancing hepatic lesions (0.5-2 cm in diameter), ie, 44 HCCs and 53 AEPs, detected on gadoxetic acid-enhanced MRI, were included in this study. HCCs were diagnosed either through histopathology confirmation (n = 16) or by a combination of liver computed tomography (CT), angiographic findings, lipiodol CT, and AFP levels (n = 28). AEPs were diagnosed either through histopathology confirmation (n = 2) or were based on the angiographic findings, liver CT, and follow-up imaging (n = 51). Two radiologists jointly analyzed the morphologic features and the enhancement characteristics on the gadoxetic acid-enhanced MRI. Of the 69 study patients, 42 patients with 60 arterial enhancing lesions underwent quadruple-phase CT in addition to their MRI examination within 4 weeks before or after the MRI, and 2 other radiologists who were blinded to the final diagnosis independently reviewed the MRI and CT images in random order, at an interval of 2 weeks. Diagnostic performance was evaluated using receiver operating characteristics. The Kappa test was used to evaluate interobserver agreement. RESULTS: Among 44 HCCs, 42 (95.4%) demonstrated low signal intensity (SI) and only 2 showed iso- or high SI on the hepatobiliary phase of gadoxetic acid-enhanced MRI. Alternatively, most AEPs showed iso SI on the hepatobiliary (n = 50, 94.3%) phase, and only 2 AEPs showed low SI. Compared with the diagnostic performance of the 2 imaging modalities, the mean areas under the receiver-operator characteristic curves on MR imaging were 0.975 for reviewer 1 and 0.966 for reviewer 2, whereas those of CT imaging were 0.892 for reviewer 1 and 0.888 for reviewer 2 (P = 0.069 and P = 0.106, respectively). The sensitivity for each reviewer with MR imaging (93.9% and 90.9%, respectively) was significantly higher than that with multiphasic CT (54.5%, in both) (P = 0.001 and 0.0018, respectively). CONCLUSION: HCCs and AEPs show different enhancing features on the delayed dynamic and hepatobiliary phases of gadoxetic acid-enhanced MRI. Gadoxetic acid-enhanced MRI may, therefore, help to differentiate between HCC and AEP.