ABSTRACT
Ovarian cancer, a group of heterogeneous diseases, presents with extensive characteristics with the highest mortality among gynecological malignancies. Accurate and early diagnosis of ovarian cancer is of great significance. Here, we present OvcaFinder, an interpretable model constructed from ultrasound images-based deep learning (DL) predictions, Ovarian-Adnexal Reporting and Data System scores from radiologists, and routine clinical variables. OvcaFinder outperforms the clinical model and the DL model with area under the curves (AUCs) of 0.978, and 0.947 in the internal and external test datasets, respectively. OvcaFinder assistance led to improved AUCs of radiologists and inter-reader agreement. The average AUCs were improved from 0.927 to 0.977 and from 0.904 to 0.941, and the false positive rates were decreased by 13.4% and 8.3% in the internal and external test datasets, respectively. This highlights the potential of OvcaFinder to improve the diagnostic accuracy, and consistency of radiologists in identifying ovarian cancer.
Subject(s)
Ovarian Neoplasms , Female , Humans , Ovarian Neoplasms/diagnostic imaging , Area Under Curve , Extremities , Radiologists , Retrospective StudiesABSTRACT
Background Use of contrast material-enhanced (CE) US Liver Imaging Reporting and Data System (LI-RADS) version 2017 has not been validated in large populations where hepatitis B virus (HBV) is endemic. Purpose To evaluate the diagnostic performance of CE US LI-RADS version 2017 in a population with a high prevalence of HBV infection. Materials and Methods In this retrospective study, liver nodules in patients with HBV who were evaluated from January 2004 to December 2016 were categorized as CE US LR-1 to LR-5 through LR-M. A subgroup of LR-M nodules was reclassified as LR-5, and additional analysis was performed. The reference standard consisted of histologic evaluation or composite imaging and clinical follow-up findings. Diagnostic performance was assessed with sensitivity, specificity, positive predictive value (PPV), and negative predictive value. Results A total of 2020 nodules in 1826 patients (median age, 54 years ± 12 [standard deviation]; 1642 men) were included. Of the 1159 LR-5 lesions, 1141 were hepatocellular carcinoma (HCC); three, intrahepatic cholangiocarcinomas; six, other malignancies; six, atypical hyperplasia; and three, benign lesions. The PPV of LR-5 for HCC was 98% (95% confidence interval [CI]: 98%, 99%). In LR-M nodules, 153 showed arterial phase hyperenhancement, early washout, and absence of punched-out appearance within 5 minutes, and 142 of 153 (93%; 95% CI: 89%, 97%) were HCC. If these nodules were reclassified as LR-5, LR-M specificity and PPV as a predictor of non-HCC malignancy increased from 88% (95% CI: 87%, 89%) and 36% (95% CI: 31%, 41%) to 96% (95% CI: 95%, 97%) and 58% (95% CI: 51%, 65%), respectively (P < .001). Despite reclassification, LR-5 specificity and PPV remained high (94% [95% CI: 92%, 96%] and 98% [95% CI: 97%, 99%], respectively). Conclusion The contrast-enhanced US Liver Imaging Reporting and Data System version 2017 category LR-5 is effectively predictive of the presence of hepatocellular carcinoma. In patients with hepatitis B virus infection, performance may be further improved by reclassification of category LR-M nodules with arterial phase hyperenhancement, early washout, and no punched-out appearance to LR-5. Published under a CC BY 4.0 license. Online supplemental material is available for this article. See also the editorial by Sidhu in this issue.