Targetoid appearance on T2-weighted imaging and signs of tumor vascular involvement: diagnostic value for differentiating HCC from other primary liver carcinomas
Cannella, R., Fraum, T.J., Ludwig, D.R. et al. Eur Radiol (2021).
Available from: https://link.springer.com/article/10.1007/s00330-021-07743-x
Keywords: Hepatocellular carcinoma, Intrahepatic cholangiocarcinoma, Magnetic resonance imaging.
Clinical question: Can targetoid appearance on T2WI and signs of tumor vascular involvement be used as potential LI-RADS features to differentiate HCC from non-HCC malignancies.
What was done: Targetoid appearance and signs of tumor vascular invasion were compared in intrahepatic cholangiocarcinomas and combined hepatocellular-cholangiocarcinomas.
How was it done: Retrospective evaluation of 165 intrahepatic cholangiocarcinomas, 74 combined hepatocellular-cholangiocarcinomas and 136 HCCs (control). Targetoid appearance and five features of tumoral vascular invasion (encasement, narrowing, tethering, occlusion and obliteration) were recorded, and their sensitivities and specificities were calculated.
Findings and results: Targetoid appearance on T2WI is highly specific for non-HCC malignancies (97.3%-98.2%). Tethering and occlusion are the most specific, among features of tumor vascular invasion, for non-HCC malignancies (97.1%-100% and 99.3%, respectively). In contrast, vascular encasement and obliteration had the highest sensitivities (34.3%-37.2% and 25.5%-29.7%, respectively).
Conclusion: Targetoid appearance on T2WI and vascular tethering and occlusion demonstrated high specificity for the differentiation of non-HCC malignancies from HCC.
Implications: Targetoid appearance on T2WI maybe valuable as a potential LI-RADS imaging feature for the LR-M category.
MRI Liver Imaging Reporting and Data System (LI-RADS) Version 2018: Impact of and Reduction in Ancillary Features
van der Pol CB, Dhindsa K, Shergill R, Zha N, Ferri M, Kagoma YK, Lee SY, Satkunasingham J, Wat J, Tsai S. Am J Roentgenol. (2021).
Available from: https://www.ajronline.org/doi/10.2214/AJR.20.23031
Keywords: Hepatocellular carcinoma, Magnetic resonance imaging, Systems analysis.
Clinical question: What is the impact of LI-RADS v2018 ancillary features on MRI? And can they be reduced without hindering the accuracy of LI-RADS?
What was done: Diagnostic accuracy of LI-RADS categories was calculated with and without ancillary features. Machine learning algorithms were used to identify noncontributory ancillary features.
How was it done: Retrospective evaluation of 222 hepatic observations for the presence or absence of major and ancillary features on MRI according to LI-RADS v2018. The LI-RADS categories and their respective diagnostic accuracy with and without ancillary features were determined. Machine learning algorithms were used to identify noncontributory ancillary features, then calculate the performance metrics of LI-RADS categories based on different combinations of ancillary features.
Findings and results: The percentage of HCCs did not differ significantly when using major features alone and with application of ancillary features. Five ancillary features were identified as noncontributory: corona enhancement, nodule-in-nodule, mosaic architecture, blood products in mass and fat in mass.
Conclusion/Implications: Several ancillary features may be removed from LI-RADS v2018 for MRI without compromising its accuracy.
Deep Learning With 3D Convolutional Neural Network for Noninvasive Prediction of Microvascular Invasion in Hepatocellular Carcinoma
Zhang Y, Lv X, Qiu J, Zhang B, Zhang L, Fang J, Li M, Chen L, Wang F, Liu S, Zhang S. J Magn Reson Imaging (2021).
Available from: https://doi.org/10.1002/jmri.27538
Keywords: Hepatocellular carcinoma, Magnetic resonance imaging, Deep learning, Microvascular invasion.
Clinical question: Preoperative prediction of microvascular invasion in HCC.
How was it done: Contrast-enhanced MR studies from 237 patients with pathologically confirmed HCC after surgical resection were used as input for four deep learning models. Three models were based on a single MR sequence, and a fourth “fusion” model combining the three MR sequences. These models were used to predict the presence or absence of microvascular invasion.
Findings and results: The highest predictive accuracy was achieved by the fusion model; 71%, with sensitivity of 55% and specificity of 81% on the validation dataset.
Conclusion/Implications: Deep learning may be used for preoperative prediction of microvascular invasion in patients with HCC.
Automated Analysis of Multiparametric Magnetic Resonance Imaging/Magnetic Resonance Elastography Exams for Prediction of Nonalcoholic Steatohepatitis
Dzyubak B, Li J, Chen J, Mara KC, Therneau TM, Venkatesh SK, Ehman RL, Allen AM, Yin M. J Magn Reson Imaging (2021).
Available from: https://doi.org/10.1002/jmri.27549
Keywords: MRE, NASH, PDFF, liver stiffness.
Clinical question: Automated analysis of steatosis and liver stiffness, and prediction of NASH on MR.
What was done: Manual analysis of proton density fat fraction (PDFF) and liver stiffness on MR studies was compared with fully automated machine learning based analysis. Another model was developed to predict NASH.
How was it done: Prospective evaluation of 83 adult patients with obesity with MR elastography (MRE) and chemical shift MR (CS-MRI). All patients had liver biopsy for assessment of NASH. PDFF and liver stiffness were manually calculated by two radiologists. An automated algorithm was developed for automated region-of-interest selection and calculation of PDFF and liver stiffness. Another model was created to predict pathology-diagnosed NASH based on stiffness and PDFF.
Findings and results: The agreement between the automated measurements and the readers was high (R2 = 0.87 for stiffness and R2 = 0.99 for PDFF). Area under the ROC curve for the NASH prediction model was 0.87.
Conclusion/Implications: NASH may be predicted using an automated machine-learning algorithm based on liver stiffness and PDFF on MRE and CS-MRI, respectively.References