• Korean Journal of Radiology
• /
• v.22 no.7
• /
• pp.1213-1224
• /
• 2021

Objective: To develop a machine learning (ML) pipeline based on radiomics to predict Coronavirus Disease 2019 (COVID-19) severity and the future deterioration to critical illness using CT and clinical variables. Materials and Methods: Clinical data were collected from 981 patients from a multi-institutional international cohort with real-time polymerase chain reaction-confirmed COVID-19. Radiomics features were extracted from chest CT of the patients. The data of the cohort were randomly divided into training, validation, and test sets using a 7:1:2 ratio. A ML pipeline consisting of a model to predict severity and time-to-event model to predict progression to critical illness were trained on radiomics features and clinical variables. The receiver operating characteristic area under the curve (ROC-AUC), concordance index (C-index), and time-dependent ROC-AUC were calculated to determine model performance, which was compared with consensus CT severity scores obtained by visual interpretation by radiologists. Results: Among 981 patients with confirmed COVID-19, 274 patients developed critical illness. Radiomics features and clinical variables resulted in the best performance for the prediction of disease severity with a highest test ROC-AUC of 0.76 compared with 0.70 (0.76 vs. 0.70, p = 0.023) for visual CT severity score and clinical variables. The progression prediction model achieved a test C-index of 0.868 when it was based on the combination of CT radiomics and clinical variables compared with 0.767 when based on CT radiomics features alone (p < 0.001), 0.847 when based on clinical variables alone (p = 0.110), and 0.860 when based on the combination of visual CT severity scores and clinical variables (p = 0.549). Furthermore, the model based on the combination of CT radiomics and clinical variables achieved time-dependent ROC-AUCs of 0.897, 0.933, and 0.927 for the prediction of progression risks at 3, 5 and 7 days, respectively. Conclusion: CT radiomics features combined with clinical variables were predictive of COVID-19 severity and progression to critical illness with fairly high accuracy.

• Korean Journal of Radiology
• /
• v.22 no.3
• /
• pp.334-343
• /
• 2021

Objective: We aimed to develop a prediction model for diagnosing severe aortic stenosis (AS) using computed tomography (CT) radiomics features of aortic valve calcium (AVC) and machine learning (ML) algorithms. Materials and Methods: We retrospectively enrolled 408 patients who underwent cardiac CT between March 2010 and August 2017 and had echocardiographic examinations (240 patients with severe AS on echocardiography [the severe AS group] and 168 patients without severe AS [the non-severe AS group]). Data were divided into a training set (312 patients) and a validation set (96 patients). Using non-contrast-enhanced cardiac CT scans, AVC was segmented, and 128 radiomics features for AVC were extracted. After feature selection was performed with three ML algorithms (least absolute shrinkage and selection operator [LASSO], random forests [RFs], and eXtreme Gradient Boosting [XGBoost]), model classifiers for diagnosing severe AS on echocardiography were developed in combination with three different model classifier methods (logistic regression, RF, and XGBoost). The performance (c-index) of each radiomics prediction model was compared with predictions based on AVC volume and score. Results: The radiomics scores derived from LASSO were significantly different between the severe AS and non-severe AS groups in the validation set (median, 1.563 vs. 0.197, respectively, p < 0.001). A radiomics prediction model based on feature selection by LASSO + model classifier by XGBoost showed the highest c-index of 0.921 (95% confidence interval [CI], 0.869-0.973) in the validation set. Compared to prediction models based on AVC volume and score (c-indexes of 0.894 [95% CI, 0.815-0.948] and 0.899 [95% CI, 0.820-0.951], respectively), eight and three of the nine radiomics prediction models showed higher discrimination abilities for severe AS. However, the differences were not statistically significant (p > 0.05 for all). Conclusion: Models based on the radiomics features of AVC and ML algorithms may perform well for diagnosing severe AS, but the added value compared to AVC volume and score should be investigated further.

• Korean Journal of Radiology
• /
• v.22 no.6
• /
• pp.959-969
• /
• 2021

Objective: This study aimed to evaluate the role of preoperative two-dimensional (2D) shear wave elastography (SWE) in assessing the stages of liver fibrosis in patients with suspected biliary atresia (BA) and compared its diagnostic performance with those of serum fibrosis biomarkers. Materials and Methods: This study was approved by the ethical committee, and written informed parental consent was obtained. Two hundred and sixteen patients were prospectively enrolled between January 2012 and October 2018. The 2D SWE measurements of 69 patients have been previously reported. 2D SWE measurements, serum fibrosis biomarkers, including fibrotic markers and biochemical test results, and liver histology parameters were obtained. 2D SWE values, serum biomarkers including, aspartate aminotransferase to platelet ratio index (APRi), and other serum fibrotic markers were correlated with the stages of liver fibrosis by METAVIR. Receiver operating characteristic (ROC) curves and area under the ROC (AUROC) curve analyses were used. Results: The correlation coefficient of 2D SWE value in correlation with the stages of liver fibrosis was 0.789 (p < 0.001). The cut-off values of 2D SWE were calculated as 9.1 kPa for F1, 11.6 kPa for F2, 13.0 kPa for F3, and 15.7 kPa for F4. The AUROCs of 2D SWE in the determination of the stages of liver fibrosis ranged from 0.869 to 0.941. The sensitivity and negative predictive value of 2D SWE in the diagnosis of ≥ F3 was 93.4% and 96.0%, respectively. The diagnostic performance of 2D SWE was superior to that of APRi and other serum fibrotic markers in predicting severe fibrosis and cirrhosis (all p < 0.005) and other serum biomarkers. Multivariate analysis showed that the 2D SWE value was the only statistically significant parameter for predicting liver fibrosis. Conclusion: 2D SWE is a more effective non-invasive tool for predicting the stage of liver fibrosis in patients with suspected BA, compared with serum fibrosis biomarkers.

• Korean Journal of Radiology
• /
• v.21 no.4
• /
• pp.413-421
• /
• 2020

Objective: A widely applicable, non-invasive screening method for non-alcoholic fatty liver disease (NAFLD) is needed. We aimed to develop and validate an index combining computed tomography (CT) and routine clinical data for screening for NAFLD in a large cohort of adults with pathologically proven NAFLD. Materials and Methods: This retrospective study included 2218 living liver donors who had undergone liver biopsy and CT within a span of 3 days. Donors were randomized 2:1 into development and test cohorts. CTL-S was measured by subtracting splenic attenuation from hepatic attenuation on non-enhanced CT. Multivariable logistic regression analysis of the development cohort was utilized to develop a clinical-CT index predicting pathologically proven NAFLD. The diagnostic performance was evaluated by analyzing the areas under the receiver operating characteristic curve (AUC). The cutoffs for the clinical-CT index were determined for 90% sensitivity and 90% specificity in the development cohort, and their diagnostic performance was evaluated in the test cohort. Results: The clinical-CT index included CTL-S, body mass index, and aspartate transaminase and triglyceride concentrations. In the test cohort, the clinical-CT index (AUC, 0.81) outperformed CTL-S (0.74; p < 0.001) and clinical indices (0.73-0.75; p < 0.001) in diagnosing NAFLD. A cutoff of ≥ 46 had a sensitivity of 89% and a specificity of 41%, whereas a cutoff of ≥ 56.5 had a sensitivity of 57% and a specificity of 89%. Conclusion: The clinical-CT index is more accurate than CTL-S and clinical indices alone for the diagnosis of NAFLD and may be clinically useful in screening for NAFLD.

• Korean Journal of Radiology
• /
• v.21 no.1
• /
• pp.88-100
• /
• 2020

Objective: We aimed to develop and validate a deep learning system for fully automated segmentation of abdominal muscle and fat areas on computed tomography (CT) images. Materials and Methods: A fully convolutional network-based segmentation system was developed using a training dataset of 883 CT scans from 467 subjects. Axial CT images obtained at the inferior endplate level of the 3rd lumbar vertebra were used for the analysis. Manually drawn segmentation maps of the skeletal muscle, visceral fat, and subcutaneous fat were created to serve as ground truth data. The performance of the fully convolutional network-based segmentation system was evaluated using the Dice similarity coefficient and cross-sectional area error, for both a separate internal validation dataset (426 CT scans from 308 subjects) and an external validation dataset (171 CT scans from 171 subjects from two outside hospitals). Results: The mean Dice similarity coefficients for muscle, subcutaneous fat, and visceral fat were high for both the internal (0.96, 0.97, and 0.97, respectively) and external (0.97, 0.97, and 0.97, respectively) validation datasets, while the mean cross-sectional area errors for muscle, subcutaneous fat, and visceral fat were low for both internal (2.1%, 3.8%, and 1.8%, respectively) and external (2.7%, 4.6%, and 2.3%, respectively) validation datasets. Conclusion: The fully convolutional network-based segmentation system exhibited high performance and accuracy in the automatic segmentation of abdominal muscle and fat on CT images.

• Korean Journal of Radiology
• /
• v.22 no.7
• /
• pp.1100-1109
• /
• 2021

Objective: To compare the diagnostic performance between Prostate Imaging-Reporting and Data System version 2.0 (PI-RADSv2.0) and version 2.1 (PI-RADSv2.1) for clinically significant prostate cancer (csPCa) in the peripheral zone (PZ). Materials and Methods: This retrospective study included 317 patients who underwent multiparametric magnetic resonance imaging and targeted biopsy for PZ lesions. Definition of csPCa was International Society of Urologic Pathology grade ≥ 2 cancer. Area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy for csPCa were analyzed by two readers. The cancer detection rate (CDR) for csPCa was investigated according to the PI-RADS categories. Results: AUC of PI-RADSv2.1 (0.856 and 0.858 for reader 1 and 2 respectively) was higher than that of PI-RADSv2.0 (0.795 and 0.747 for reader 1 and 2 respectively) (both p < 0.001). Sensitivity, specificity, PPV, NPV, and accuracy for PI-RADSv2.0 vs. PI-RADSv2.1 were 93.2% vs. 88.3% (p = 0.023), 52.8% vs. 76.6% (p < 0.001), 48.7% vs. 64.5% (p < 0.001), 94.2% vs. 93.2% (p = 0.504), and 65.9% vs. 80.4% (p < 0.001) for reader 1, and 96.1% vs. 92.2% (p = 0.046), 34.1% vs. 72.4% (p < 0.001), 41.3% vs. 61.7% (p < 0.001), 94.8% vs. 95.1% (p = 0.869), and 54.3% vs. 78.9% (p < 0.001) for reader 2, respectively. CDRs of PI-RADS categories 1-2, 3, 4, and 5 for PI-RADSv2.0 vs. PI-RADSv2.1 were 5.9% vs. 5.9%, 5.8% vs. 12.5%, 39.8% vs. 56.2%, and 88.9% vs. 88.9% for reader 1; and 4.5% vs. 4.1%, 6.1% vs. 11.1%, 32.5% vs. 53.4%, and 85.0% vs. 86.8% for reader 2, respectively. Conclusion: Our data demonstrated improved AUC, specificity, PPV, accuracy, and CDRs of category 3 or 4 of PI-RADSv2.1, but decreased sensitivity, compared with PI-RADSv2.0, for csPCa in PZ.

• Korean Journal of Radiology
• /
• v.22 no.8
• /
• pp.1240-1252
• /
• 2021

Objective: To compare the accuracy for detecting breast cancer in the diagnostic setting between the use of digital breast tomosynthesis (DBT), defined as DBT alone or combined DBT and digital mammography (DM), and the use of DM alone through a systematic review and meta-analysis. Materials and Methods: Ovid-MEDLINE, Ovid-Embase, Cochrane Library and five Korean local databases were searched for articles published until March 25, 2020. We selected studies that reported diagnostic accuracy in women who were recalled after screening or symptomatic. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. A bivariate random effects model was used to estimate pooled sensitivity and specificity. We compared the diagnostic accuracy between DBT and DM alone using meta-regression and subgroup analyses by modality of intervention, country, existence of calcifications, breast density, Breast Imaging Reporting and Data System category threshold, study design, protocol for participant sampling, sample size, reason for diagnostic examination, and number of readers who interpreted the studies. Results: Twenty studies (n = 44513) that compared DBT and DM alone were included. The pooled sensitivity and specificity were 0.90 (95% confidence interval [CI] 0.86-0.93) and 0.90 (95% CI 0.84-0.94), respectively, for DBT, which were higher than 0.76 (95% CI 0.68-0.83) and 0.83 (95% CI 0.73-0.89), respectively, for DM alone (p < 0.001). The area under the summary receiver operating characteristics curve was 0.95 (95% CI 0.93-0.97) for DBT and 0.86 (95% CI 0.82-0.88) for DM alone. The higher sensitivity and specificity of DBT than DM alone were consistently noted in most subgroup and meta-regression analyses. Conclusion: Use of DBT was more accurate than DM alone for the diagnosis of breast cancer. Women with clinical symptoms or abnormal screening findings could be more effectively evaluated for breast cancer using DBT, which has a superior diagnostic performance compared to DM alone.

• Korean Journal of Radiology
• /
• v.22 no.5
• /
• pp.829-839
• /
• 2021

Objective: To compare the diagnostic performance of contrast-enhanced radial T1-weighted gradient-echo 3-tesla (3T) magnetic resonance imaging (MRI) and computed tomography (CT) for the detection of visceral pleural surface invasion (VPSI). Visceral pleural invasion by non-small-cell lung cancer (NSCLC) can be classified into two types: PL1 (without VPSI), invasion of the elastic layer of the visceral pleura without reaching the visceral pleural surface, and PL2 (with VPSI), full invasion of the visceral pleura. Materials and Methods: Thirty-three patients with pathologically confirmed VPSI by NSCLC were retrospectively reviewed. Multidetector CT and contrast-enhanced 3T MRI with a free-breathing radial three-dimensional fat-suppressed volumetric interpolated breath-hold examination (VIBE) pulse sequence were compared in terms of the length of contact, angle of mass margin, and arch distance-to-maximum tumor diameter ratio. Supplemental evaluation of the tumor-pleura interface (smooth versus irregular) could only be performed with MRI (not discernible on CT). Results: At the tumor-pleura interface, radial VIBE MRI revealed a smooth margin in 20 of 21 patients without VPSI and an irregular margin in 10 of 12 patients with VPSI, yielding an accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F-score for VPSI detection of 91%, 83%, 95%, 91%, 91%, and 87%, respectively. The McNemar test and receiver operating characteristics curve analysis revealed no significant differences between the diagnostic accuracies of CT and MRI for evaluating the contact length, angle of mass margin, or arch distance-to-maximum tumor diameter ratio as predictors of VPSI. Conclusion: The diagnostic performance of contrast-enhanced radial T1-weighted gradient-echo 3T MRI and CT were equal in terms of the contact length, angle of mass margin, and arch distance-to-maximum tumor diameter ratio. The advantage of MRI is its clear depiction of the tumor-pleura interface margin, facilitating VPSI detection.

• Korean Journal of Radiology
• /
• v.22 no.1
• /
• pp.106-117
• /
• 2021

Objective: To explore the value of magnetic resonance imaging (MRI)-based whole tumor texture analysis in differentiating borderline epithelial ovarian tumors (BEOTs) from FIGO stage I/II malignant epithelial ovarian tumors (MEOTs). Materials and Methods: A total of 88 patients with histopathologically confirmed ovarian epithelial tumors after surgical resection, including 30 BEOT and 58 MEOT patients, were divided into a training group (n = 62) and a test group (n = 26). The clinical and conventional MRI features were retrospectively reviewed. The texture features of tumors, based on T2-weighted imaging, diffusion-weighted imaging, and contrast-enhanced T1-weighted imaging, were extracted using MaZda software and the three top weighted texture features were selected by using the Random Forest algorithm. A non-texture logistic regression model in the training group was built to include those clinical and conventional MRI variables with p value < 0.10. Subsequently, a combined model integrating non-texture information and texture features was built for the training group. The model, evaluated using patients in the training group, was then applied to patients in the test group. Finally, receiver operating characteristic (ROC) curves were used to assess the diagnostic performance of the models. Results: The combined model showed superior performance in categorizing BEOTs and MEOTs (sensitivity, 92.5%; specificity, 86.4%; accuracy, 90.3%; area under the ROC curve [AUC], 0.962) than the non-texture model (sensitivity, 78.3%; specificity, 84.6%; accuracy, 82.3%; AUC, 0.818). The AUCs were statistically different (p value = 0.038). In the test group, the AUCs, sensitivity, specificity, and accuracy were 0.840, 73.3%, 90.1%, and 80.8% when the non-texture model was used and 0.896, 75.0%, 94.0%, and 88.5% when the combined model was used. Conclusion: MRI-based texture features combined with clinical and conventional MRI features may assist in differentitating between BEOT and FIGO stage I/II MEOT patients.

• Korean Journal of Radiology
• /
• v.22 no.2
• /
• pp.253-262
• /
• 2021

Objective: To determine whether the values of hepatic apparent diffusion coefficient (ADC) can differentiate biliary atresia (BA) from non-BA or be correlated with the grade of hepatic fibrosis in infants with cholestasis. Materials and Methods: This retrospective cohort study included infants who received liver MRI examinations to evaluate cholestasis from July 2009 to October 2017. Liver ADC, ADC ratio of liver/spleen, aspartate aminotransferase to platelet ratio index (APRI), and spleen size were compared between the BA and non-BA groups. The diagnostic performances of all parameters for significant fibrosis (F3-4) were obtained by receiver-operating characteristics (ROCs) curve analysis. Results: Altogether, 227 infants (98 males and 129 females, mean age = 57.2 ± 36.3 days) including 125 BA patients were analyzed. The absolute ADC difference between two reviewers was 0.10 mm²/s for both liver and spleen. Liver ADC value was specific (80.4%) and ADC ratio was sensitive (88.0%) for the diagnosis of BA with comparable performance. There were 33 patients with F0, 15 with F1, 71 with F2, 35 with F3, and 11 with F4. All four parameters of APRI (τ = 0.296), spleen size (τ = 0.312), liver ADC (τ = -0.206), and ADC ratio (τ = -0.288) showed significant correlation with fibrosis grade (all, p < 0.001). The cutoff values for significant fibrosis (F3-4) were 0.783 for APRI (area under the ROC curve [AUC], 0.721), 5.9 cm for spleen size (AUC, 0.719), 1.044 x 10^-3 mm²/s for liver ADC (AUC, 0.673), and 1.22 for ADC ratio (AUC, 0.651). Conclusion: Liver ADC values and ADC ratio of liver/spleen showed limited additional diagnostic performance for differentiating BA from non-BA and predicting significant hepatic fibrosis in infants with cholestasis.