Radiomics analysis of [18F] FDG PET/CT for microvascular invasion and prognosis prediction in very-early and early-stage hepatocellular carcinoma

Abstract

We have read with keen interest the article of Youcai Li and colleagues, published in the European Journal of Nuclear Medicine and Molecular Imaging titled “Radiomics analysis of [18F] FDG positron emission tomography/computed tomography (PET/CT) for microvascular invasion and prognosis prediction in very-early and early-stage hepatocellular carcinoma” [1]. We appreciate the valuable insights of the authors for developing a radiomics nomogram of PET/CT to predict microvascular invasion (MVI) status and disease-free survival (DFS) in patients with hepatocellular carcinoma (HCC) at very-early stage and early-stage of the disease (BCLC 0-A), which could enable a step forward in precision medicine. To the best of our knowledge, no study has been conducted on the radiomics analysis to predict MVI and prognosis in very early and early stages of HCC, and the importance of timely and effective treatment of patients. However, we would like to address several concerns. First, in part of the “Development of individualized prediction models,” patients were classified into the high and low-risk groups by the “X-tile” software for the radiomics signature. We would like to know why the authors did not use Kaplan–Meier (KM) survival curve for analyzing patient survival outcomes. According to previous studies, the “X-tile” software is usually used to determine the optimal cut-off values for survival analysis [2, 3]; KM analyses and log-rank tests are usually performed to identify survival differences. Second, although the calibration curve proves the stability of the prediction model in training cohort, recent studies reported that the prediction error of models could be assessed by the “Boot632plus” split method to calculate the Brier score, which reflects a weighted average of the squared distance between an observed recurrence status and predicted recurrence probability of a model; this may be more convincing than the calibration curve valid model performance [4]. To predict the MVI status of early-stage HCC, binary logistic regression analyses were performed to select the significant variables from the training cohort. Therefore, the description in Table 2, “Univariate cox regression analysis was used for MVI in the training cohort,” could be probably changed in “Univariate logistic regression analysis was used for MVI in the training cohort.” Third, one cannot deny the potential of radiomics as an alternative to histology and ignore the clinical need for radiomics. However, manual analysis requires considerable time and effort. In addition, the clinical application of radiomics is affected by the equipment of different manufacturers. Consequently, a nomogram based on PET/CT still needs to be validated in the future. We look forward to positive responses to our concerns.