Abstract
ObjectiveThe aim of this study was to predict the prognosis in patients with metastatic rectal cancer (mRC) by obtaining a model with machine learning (ML) algorithms through volumetric and radiomic data obtained from baseline 18-Fluorine Fluorodeoxyglucose ([18F]FDG) positron emission tomography/computed tomography (PET/CT) images. MethodsSixty-two patients with mRC who underwent [18F]FDG PET/CT imaging for staging between January 2015 and January 2021 were evaluated using LIFEx software. The volume of interest (VOI) of the primary tumor was generated and volumetric and textural features were obtained from this VOI. In addition, the total metabolic tumor volume (tMTV) and total lesion glycolysis (TLG) values of tumor foci in the whole body were evaluated. Clinical and radiomic data were evaluated with ML algorithms to create a model that predicts survival. Significant associations between these features and 1- and 2-year survival were investigated. ResultsThe random forest algorithm was the most successful in predicting 2-year survival (AUC: 0.843, precision-recall curve: 0.822 and Matthew's correlation coefficient: 0.583). The model obtained with this algorithm was able to predict 49 patients with 79.03% accuracy. While tMTV and TLG values were successful in predicting 1-year survival (p: 0.002 and 0.007, respectively), texture characteristics of the primary tumor did not show a significant relationship with 1-year survival. ConclusionsIn addition to the important role of [18F]FDG PET/CT in staging patients with mRC, this study shows that it is possible to predict survival with ML methods, with parameters obtained using texture analysis of the primary tumor and whole body volumetric parameters.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have