23 April 2024 in Liver International
Background and Aims: The Liver Imaging Reporting and Data System (LI‐RADS) offers a standardized approach for imaging hepatocellular carcinoma. However, the diverse styles and structures of radiology reports complicate automatic data extraction. Large language models hold the potential for structured data extraction from free‐text reports. Our objective was to evaluate the performance of Generative Pre‐trained Transformer (GPT)‐4 in extracting LI‐RADS features and categories from free‐text liver magnetic resonance imaging (MRI) reports.Methods: Three radiologists generated 160 fictitious free‐text liver MRI reports written in Korean and English, simulating real‐world practice. Of these, 20 were used for prompt engineering, and 140 formed the internal test cohort. Seventy‐two genuine reports, authored by 17 radiologists were collected and de‐identified for the external test cohort. LI‐RADS features were extracted using GPT‐4, with a Python script calculating categories. Accuracies in each test cohort were compared.Results: On the external test, the accuracy for the extraction of major LI‐RADS features, which encompass size, nonrim arterial phase hyperenhancement, nonperipheral ‘washout’, enhancing ‘capsule’ and threshold growth, ranged from .92 to .99. For the rest of the LI‐RADS features, the accuracy ranged from .86 to .97. For the LI‐RADS category, the model showed an accuracy of .85 (95% CI: .76, .93).Conclusions: GPT‐4 shows promise in extracting LI‐RADS features, yet further refinement of its prompting strategy and advancements in its neural network architecture are crucial for reliable use in processing complex real‐world MRI reports.
1 October 2021 in Ultrasonography
Purpose: The aim of this study was to evaluate the association of contrast-enhanced ultrasound (CEUS) features using Sonazoid for liver nodules with Liver Imaging Reporting and Data System (LI-RADS) categories and to identify the usefulness of Kupffer-phase images.Methods: This retrospective study was conducted in 203 patients at high risk of hepatocellular carcinoma (HCC) who underwent CEUS with Sonazoid from 2013 to 2016. Nodule enhancement in the arterial, portal venous, late, and Kupffer phases; CEUS LI-RADS major features; and Kupffer-phase defects were evaluated. According to the computed tomography/magnetic resonance imaging (CT/MRI) LI-RADS v2018, all nodules were assigned an LR category (n=4/33/99/67 for LR-M/3/4/5) and comparisons across LR categories were made. We defined modified CEUS LI-RADS as using Kupffer-phase defects as an alternative to late and mild washout in CEUS LI-RADS and compared the diagnostic performance for HCC.Results: On CEUS of 203 nodules, 89.6% of CT/MRI LR-5 and 85.9% of LR-4 nodules showed hyperenhancement in the arterial phase, while 57.6% of LR-3 nodules showed hyperenhancement. Among the CT/MRI LR-5 nodules that showed arterial phase hyperenhancement or isoenhancement, 59.7% showed hypoenhancing changes from the portal venous phase, 23.9% from the late phase, and 13.4% additionally in the Kupffer phase. The modified CEUS LI-RADS showed higher sensitivity than CEUS LI-RADS (83.2% vs. 74.2%, P=0.008) without compromising specificity (63.6% vs. 69.7%, P=0.500).Conclusion: The Kupffer phase best shows hypoenhancing changes in LR-5 lesions and is expected to improve the sensitivity for HCC in high-risk patients.
1 August 2023 in Radiology
Background Ovarian-Adnexal Reporting and Data System (O-RADS) US provides a standardized method with which to stratify lesions into risk of malignancy categories, which is crucial for proper management. Purpose To perform a systematic review and meta-analysis to estimate malignancy rates for each O-RADS US score and evaluate the diagnostic performance of combined O-RADS US scores 4 and 5 in the diagnosis of malignancy. Materials and Methods A systematic literature search from the inception of the MEDLINE, EMBASE, and Web of Science databases through January 27, 2023, was performed for articles that reported using the O-RADS US stratification system and included malignancy rates per each O-RADS score. Bivariate random-effects models were used to determine the pooled malignancy rates for each O-RADS US score and to obtain summary estimates of the diagnostic performance of combined O-RADS US scores 4 and 5 in the diagnosis of malignant lesions. Results The final analysis included 18 studies consisting of 11 605 patients and 11 818 ovarian-adnexal lesions, with 2996 malignant (25.4%) and 8822 benign (74.6%) lesions. No malignant lesions were reported in O-RADS 1 category. The pooled percentages of malignancy were 0.6% (95% CI: 0.3, 1.0) for O-RADS 2, 3.9% (95% CI: 2.5, 5.4) for O-RADS 3, 43.5% (95% CI: 33.8, 53.2) for O-RADS 4, and 87.3% (95% CI: 83.0, 91.7) for O-RADS 5. The pooled sensitivity and specificity of combined O-RADS scores 4 and 5 in the diagnosis of malignant lesions were 95.6% (95% CI: 94.0, 97.2) and 76.6% (95% CI: 70.4, 82.7), respectively. Conclusion Each O-RADS US score provided the intended probability of malignant lesions as outlined by the O-RADS risk stratification system. When O-RADS US scores 4 and 5 were combined as a predictor for malignancy, O-RADS US showed a high sensitivity and moderate specificity. Clinical trial registration no. CRD42022352166 © RSNA, 2023 Supplemental material is available for this article.
1 January 2023 in Korean Journal of Radiology
Jeong WK, et al. Korean J Radiol. 2023 Jun;24(6):482-497. https://doi.org/10.3348/kjr.2023.0324
1 January 2021 in Korean Journal of Radiology
Moon JY, et al. Korean J Radiol. 2021 Jul;22:e98. https://doi.org/10.3348/kjr.2020.1428