Preoperative prediction of lymph node metastasis in colorectal cancer using 18F‐FDG PET/CT peritumoral radiomics analysis
Yan Wang,Hongyue Zhao,Peng Fu,Lin Tian,Yexin Su,Zhehao Lyu,Wenchao Gu,Yang Wang,Shan Liu,Xi Wang,Han Zheng,Jingjing Du,Rui Zhang
DOI: https://doi.org/10.1002/mp.17193
IF: 4.506
2024-05-29
Medical Physics
Abstract:Background Radiomics has been used in the diagnosis of tumor lymph node metastasis (LNM). However, to date, most studies have been based on intratumoral radiomics. Few studies have focused on the use of 18F‐fluorodeoxyglucose positron emission computed tomography (18F‐FDG PET/CT) peritumoral radiomics for the diagnosis of LNM in colorectal cancer (CRC). Purpose Determining the value of radiomics features extracted from 18F‐FDG PET/CT images of the peritumoral region in predicting LNM in patients with CRC. Methods The clinical data and preoperative 18F‐FDG PET/CT images of 244 CRC patients were retrospectively analyzed. Intratumoral and peritumoral radiomics features were screened using the mutual information method, and least absolute shrinkage and selection operator regression. Based on the selected radiomics features, a radiomics score (Rad‐score) was calculated, and independent risk factors obtained from univariate and multivariate logistic regression analyses were used to construct clinical and combined (Radiomics + Clinical) models. The performance of these models was evaluated using the DeLong test, while their clinical utility was assessed by decision curve analysis. Finally, a nomogram was constructed to visualize the predictive model. Results The most optimal set of features retained by the feature filtering process were all peritumoral radiomic features. Carcinoembryonic antigen levels, PET/CT‐reported lymph node status and Rad‐score were found to be independent risk factors for LNM. All three LNM risk assessment models exhibited good predictive performance, with the combined model showing the best classification results, with areas under the curve of 0.85 and 0.76 in the training and validation groups, respectively. The DeLong test revealed that the performance of the combined model was superior to that of the clinical and radiomics models in both the training and validation groups, although this difference was only statistically significant in the training group. DCA indicated that the combined model displayed better clinical utility. Conclusions 18F‐FDG PET/CT peritumoral radiomics is uniquely suited to predict the presence of LNM in patients with CRC. In particular, the predictive efficacy of LNM for precision therapy and individualized patient management can be improved by using a combination of clinical risk factors.
radiology, nuclear medicine & medical imaging