Abstract
356
Purpose: To investigate predictive factors of liver toxicity using Yttrium-90 (90Y-TARE) for unresectable hepatocellular carcinoma (uHCC) based on pretreatment 18F-FDG PET whole-liver radiomics model.
Methods: We analyzed the pretreatment 18F-FDG-PET of 39 consecutive patients with uHCC undergoing 90Y-TARE (27 resin and 12 glass microspheres). Extraction of radiomics features was performed using a whole-liver semi-automatic segmentation. Values of serum albumin, hepatic enzymes (ALT, AST), alkaline phosphatase (aP), gamma-GT (gGT) and total bilirubin performed 3 months after 90Y-TARE were collected and toxicity the respective values was graded according to the Common Terminology Criteria Adverse Events (CTCAE v4.03).
Results: We extracted a total of 108 radiomics features. Follow-up -laboratory revealed albumin toxicity in 2 patients, ALT- toxicity in 7 patients, AST toxicity in 17 patients, aP in 21 patients, gGT in 30 patients and total bilirubin toxicity in 10 patients (6 patients with grade 1 and 4 with grade 2). ROC analysis showed three whole-liver textural features (homogeneity3, inverse_difference_moment2 and max_spectrum) to be able to predict total bilirubin toxicity at 3 months; AUC 0.728 (CI 0.569-0.886, p=0.034), AUC 0.728 (CI 0.570-0.885, p=0.034) and AUC 0.734 (CI 0.580-0.889, p=0.029), respectively. Whole-liver radiomics features did not predict albumin, ALT-SGPT, AST-ASAT, aP, or gGT toxicity in patients undergoing 90Y-TARE.
Conclusions: The proposed pretreatment 18F-FDG-PET whole-liver radiomics algorithm can predict bilirubin toxicity in patients undergoing 90Y-TARE for uHCC. Bilirubin is the essential factor indicating radiation induced liver disease. Our radiomics algorithm analyzes patterns of tumoral and non-tumoral liver, taking into account the relationship of uHCC and underlying cirrhosis. The proposed reader independent whole-liver algorithm might serves as prototypical model how to use FDG - PET to systemically analyze disease.