Machine learning models for predicting microvascular invasion in hepatocellular carcinoma with three-dimensional whole-lesion <sup>18</sup>F-FDG PET radiomics.
Authors
Affiliations (1)
Affiliations (1)
- Department of Nuclear Medicine, Guangxi Medical University Cancer Hospital, Nanning 530021, Guangxi Zhuang Autonomous Region, China. [email protected].
Abstract
To estimate the performance of machine learning models based on preoperative three-dimensional whole-lesion radiomics features for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC). This retrospective study included 106 individuals who underwent preoperative fluorine-18-fluorodeoxyglucose (<sup>18</sup>F-FDG) positron emission tomography/computed tomography (PET/CT) and were pathologically diagnosed with HCC at Guangxi Medical University Cancer Hospital between January 2018 and December 2023. Individuals were randomly assigned to training (n=74) and validation (n=32) sets. A total of 2,016 radiomic features were extracted from three-dimensional whole-lesion PET images. Least absolute shrinkage and selection operator regression combined with recursive feature elimination identified the optimal feature subset. Logistic regression (LR), Light Gradient Boosting Machine (LightGBM), and multilayer perceptron (MLP) models were developed. Model performance in the validation set was evaluated using precision, area under the receiver operating characteristic curve (AUC), and F1-score. Six key radiomic features were selected. In the validation set, the LR model achieved an AUC of 0.703, precision of 0.750, and F1-score of 0.765; the LightGBM model achieved an AUC of 0.654, precision of 0.688, and F1-score of 0.687; and the MLP model achieved an AUC of 0.629, precision of 0.625, and F1-score of 0.625. The LR model demonstrates the best overall performance and shows promise for noninvasive preoperative MVI assessment.