AUTHOR=Ghidaglia Jérôme , Laurent Vincent , Sebagh Mylène , Pascale Alina , Durand Emmanuel , Golse Nicolas , Besson Florent L. 

TITLE=Influence of key histological characteristics on 18F-fluorodeoxyglucose /18F-choline positron emission tomography positivity in hepatocellular carcinoma: A machine learning study

JOURNAL=Frontiers in Medicine

VOLUME=Volume 10 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2023.1087957

DOI=10.3389/fmed.2023.1087957

ISSN=2296-858X

ABSTRACT=Purpose: to decipher the influence of key-histological characteristics on the 18F-FDG and 18F-Choline PET positivity in HCC.
Material and methods: the 18F-FDG/ 18F-Choline PET imaging findings of 103 histologically-proven HCC (62 patients, 47 hepatectomies and 15 liver transplantations) were retrospectively confronted to the following key-histological parameters:  grade, capsule, micro (mVI) and macro-vascular invasion (MVI), and necrosis. Using a ratio of 70/30 for training and testing sets respectively, a penalized classification model (Elastic Net) was trained by using 100 times repeated cross validation procedures (10-fold cross-validation for hyperparameters optimization). The contribution of each histological parameters to the PET positivity was deciphered by using the Shapley Additive explanations method (SHAP). ROC curves with and without dimensionality reduction were finally estimated and compared. 
Results: Among 5 key-histological characteristics (Grade, capsule, mVI, MVI, necrosis), the mVI and tumour Grade (II-III) showed the highest relevance and robustness to explain the 18F-FDG and 18F-Choline uptakes by HCC. On the other hand, MVI and necrosis status both showed high instability in outcome predictions. Tumour capsule had very low influence on model’s predictions. By retaining only mVI and Grades II-III for the final analysis, the AUC scores were maintained (0.69 vs 0.64, 0.66 vs 0.64 and 0.64 vs 0.63 respectively for 18F-FDG, 18F-Choline and their combination).
Conclusion: a mixture of two hallmarks of aggressiveness - high grade and microvascular invasion - were the most relevant parameters to explain both the 18F-FDG and 18F-Choline PET positivity in HCC. The consideration of the tumour microenvironment will probably be necessary to improve our understanding of multitracer PET positivity in this field.