SIR 2024
Interventional Oncology
Jonathan Renslo, MS
Medical Student
USC
Financial relationships: Full list of relationships is listed on the CME information page.
Frank Yuan, MD
DR resident
USC Keck school of Medicine/ Department of Radiology
Disclosure information not submitted.
Navjit Dullet, DO
Resident Physician
University of Arizona - Tucson
Disclosure information not submitted.
Xiaomeng Lei, MPH
Statistician
Keck School of Medicine, University of Southern California
Disclosure information not submitted.
Steven Cen, PhD
Professor Of Research
Keck School of Medicine, University of Southern California
Disclosure information not submitted.
Darryl Hwang, PhD
Assistant Professor, Research Radiology
Keck School of Medicine of USC, 4D Quantitative Imaging Lab
Disclosure information not submitted.
Jenanan Vairavamurthy, MD
Assistant Professor, Interventional Radiologist
Keck School of Medicine, Univeristy of Southern California
Disclosure information not submitted.
Vinay Duddalwar, MD
Professor of Clinical Radiology
Keck School of Medicine, University of Southern California, United Kingdom
Disclosure information not submitted.
In practice, the assessment of the effectiveness of Y90 radioembolization for early stage hepatocellular carcinoma can be difficult, as the affected treatment zone often overlaps completely with the target lesions, and inflammatory effects can obscure or confound the detection of viable tumor. Radiomics may be applied to clinically acquired imaging for prognosis and outcome prediction following radioembolization therapy. This analysis of radiomic features in a cohort of Y90-treated patients assesses the utility of radiomics in making this assessment.
Materials and Methods: Four-phase CT of the abdomen from 10 patients (age 66.2±9.3 mean±stdev, 3:7 female:male) before and 1-month-post Y90 radioablation of HCC were collected. Tumor segmentations for 17 total target lesions were drawn in the venous phase using Synapse3D, then registered to other phases using whole liver registration using ANTs. Radiomics analysis produced over 3700 features consisting of 10 families of analysis: FFT2D, GLCM2D, GLCM3D, GLDM2D, GLDM3D, GLSZM2D, GLSZM3D, Intensity, LTE2D, and LTE3D. Radiological outcome was assessed at the lesion level using the mRECIST criteria on 6-month-post intervention CT. Independent t-test or Wilcoxon rank sum test was used depending on data normality to assess the difference in radiomic features between complete/partial response and stable/progressive disease.
Results: Univariate analysis of radiomics features revealed 45 features with p-value < 0.05. The top 5 most significant features are shown in Table 1. The features with strongest difference were from the signal intensity family and the texture family based on the arterial phase of the post-treatment scan.
Conclusion: Preliminary analysis of radiomics analysis indicates useful features in predicting mRECIST outcome per lesion. The strongest features come from the arterial phase of the post-treatment scan, which aligns with clinical intuition and practice. Further work with more patient data, incorporating clinical data, with more extensive modeling and analysis is under way to fully evaluate this hypothesis.