SIR 2024
Imaging
Kela Liu (she/her/hers)
Undergraduate Student
University of California, Santa Barbara
Financial relationships: Full list of relationships is listed on the CME information page.
Guangbo Yu, MS
PhD Student
University of California, Irvine
Financial relationships: Full list of relationships is listed on the CME information page.
Zigeng Zhang, MD
Postdoctoral Fellow
University of California, Irvine
Financial relationships: Full list of relationships is listed on the CME information page.
Vahid Yaghmai, n/a
Chairman of Radiology
University of California, Irvine
Disclosure information not submitted.
Aydin Eresen, PhD (he/him/his)
Postdoctoral Fellow
University of California, Irvine
Financial relationships: Full list of relationships is listed on the CME information page.
The assessment of treatment outcomes of NK cell therapy in hepatocellular carcinoma (HCC) at an early stage remains a challenge due to the absence of immediate discernible alterations in tumor size. Our study explores the feasibility of employing machine learning models based on radiomic features to achieve a more accurate appraisal of treatment outcomes of NK cell delivery and its combination with Sorafenib treatment in HCC.
Materials and Methods:
Buffalo rats were implanted with NIS1 tumors, followed by the catheterization of the proper hepatic artery in NK cell immunotherapy and combination groups. The catheterization process involved the surgical exposure of the portal triad, temporary ligation of the common hepatic artery (CHA), permanent ligation of the gastroduodenal artery (GDA), and insertion of a microcatheter from the gastroduodenal artery into the proper hepatic artery.
One week into the treatment regimen, T1W MRI data were collected with tumor regions subsequently identified. MRI features were extracted, normalized, and subjected to filtering. A two-step feature selection was performed on T1W features, including the elimination of highly correlated features and the application of the linear-kernel SVM algorithm. Binary classification models for NK vs. Control and the Combination of NK and Sorafenib vs. Control were constructed using SVM, XGBoost, and Random Forest algorithms, all with a 5-fold cross-validation. The performances were evaluated based on accuracy, areas under the curve (AUC) of the receiver operating characteristic curves (ROC), specificity, and sensitivity.
Results: For the NK cell immunotherapy group, the Random Forest model presents an accuracy of 88%, an AUC of 87%, a sensitivity of 90%, and a specificity of 83%. For the combination of NK cell and Sorafenib treatment group, the Random Forest model demonstrates better performances with an accuracy of 96%, an AUC of 100%, a sensitivity of 93%, and a specificity of 100%.
Conclusion: In general, T1W MRI radiomic features provide promising insights into the effects of NK cell therapy on HCC treatment, with a further improvement when NK therapy was combined with Sorafenib. The results underscore the potential of employing quantitative MRI analysis as a means to monitor and assess the effectiveness of NK cell therapies in HCC.