Click here to


Are you sure ?

Yes, do it No, cancel

Radiomics of Sulfur Colloid SPECT/CT to Predict Child-Pugh Class in Hepatocellular Carcinoma Patients

S Schaub1*, D Hippe1 , S Bowen1 , L Wootton1 , W Chaovalitwongse2 , P Kinahan1 , S Apisarnthanarax1 , M Nyflot1 , (1) University of Washington, Seattle, WA, (2) University of Arkansas, Fayetteville, AR


(Sunday, 7/14/2019) 4:00 PM - 5:00 PM

Room: Exhibit Hall | Forum 2

Purpose: An unmet need exists for objective biomarkers for predicting baseline liver function prior to radiotherapy in hepatocellular carcinoma (HCC) patients to minimize the risk of radiation-induced liver disease. We hypothesize that a radiomic signature derived from 99m-Tc sulfur colloid (SC) SPECT/CT could serve as a surrogate for Child-Pugh-class, the current standard for clinical assessment.

Methods: We retrospectively reviewed 92 HCC patients treated with RT for clinical data including CP score, prior liver-directed therapies (LDT), vascular invasion, and gross tumor volume (GTV). 30 radiomic features were extracted from pretreatment SC SPECT/CT of the uninvolved liver. Univariate logistic regression analysis was used to evaluate association between radiomic features and CP class (B/C vs A) following multiple comparison adjustment (α=0.001). The LASSO was used to construct a multivariate model. Model performance was reported as c-index following internal validation with the .632 bootstrap (C.632).

Results: Patients had a median age of 69 years [38-93] with 52.2% prior LDT, and 31.5% with any vascular invasion. Thirty patients (32.6%) had CP-B7+ cirrhosis and median GTV was 33 cc [1-1617]. The liver-to-spleen mean (L/Smean) ratio was strongly associated with baseline CP status (OR=0.2 per 1-SD increase, p<0.001), as were 21 radiomic texture features (p≤0.001 for all). These features remained significant after adjusting for liver volume. Conversely, sex, age, prior LDT, any vascular invasion, and GTV were not significantly associated with baseline CP status. A model using only L/Smean had C.632 = 0.83 (95% C.I. 0.75—0.92). The LASSO selected 3 radiomic features (L/Smean, skewness, LZLGE) with C.632 = 0.83 (95% C.I. 0.75—0.93), but was not significantly more accurate than L/Smean alone (p=0.98)

Conclusion: Quantitative SC SPECT features offer promise for classifying liver function as compared to Child-Pugh class. A simple model using L/Smean alone could allow for ease of adoption in clinic upon validation.

Funding Support, Disclosures, and Conflict of Interest: Dr. Nyflot reports funding from a Research Scholar grant from the Radiological Society of North America.


Not Applicable / None Entered.


IM- SPECT : Biomarkers

Contact Email