A new study published in the European Journal of Cancer has found that several radiomic features collected through magnetic resonance imaging (MRI) are useful for predicting the T stage as well as the level of perineural and microvascular invasion of extrahepatic cholangiocarcinoma (eCCA).
Different combinations of data pretreatment, dimensionality reduction, and classifiers were tested during the study, with slightly different combinations being deemed the best for each eCCA characteristic. The best radiomics model for the T stage yielded an area under the receiver operating characteristic curve (AUC) value of 1 for the model training stage and 0.962 for the test stage. The best models for perineural invasion and microvascular invasion were both able to achieve AUC values of 1 and 1 during training and testing.
The authors stated, “The results are encouraging and have the potential to become a clinically relevant tool in the management of eCCA, especially in the preoperative work-up. Multicenter retrospective and prospective validation will be implemented in subsequent studies to obtain a higher level of evidence.”
Read more about cholangiocarcinoma diagnosis
The study extracted radiomic features from T1-weighted, T2-weighted, and diffusion-weighted MRI scans as well as the apparent diffusion coefficient maps of 101 patients with eCCA. The patients had ages ranging from 28 to 83 with a mean of 56.5 and their images were collected between August 2011 and December 2019. A total of 1208 radiomics features were extracted from the 4 MRI images for each patient. Patient data were randomly assigned to a training set (80%) and a testing set (20%).
A total of 8 data pretreatment methods, 18 dimensionality reduction methods, and 12 classifiers were tested using the radiomics features in various combinations. The optimal algorithm combinations for predicting T stage, microvascular invasion, and perineural invasion were selected to create a model for each eCCA characteristic.
A combination of L1-normalization, linear discriminant analysis (LDA), and logistic regression was found to be optimal for predicting the T stage. A Box-Cox transformation combined with LDA and a K-nearest neighbor algorithm was best for predicting perineural invasion, while the combination of L2-normalization, LDA, and an adaptive boosting (AdaBoost) classifier was optimal for microvascular invasion.
Huang X, Shu J, Yan Y, et al. Feasibility of magnetic resonance imaging-based radiomics features for preoperative prediction of extrahepatic cholangiocarcinoma stage. Eur J Cancer. 2021;155:227-235.