ADVERTISEMENT
PET/CT radiomic features to predict clinical outcomes in locally advanced pancreatic cancer
Background
Innovative biomarkers to predict clinical outcomes in pancreatic cancer would be helpful in optimizing personalized treatment approaches. In this study, we aimed to develop PET/CT-based radiomic biomarkers to predict early progression in patients with locally advanced pancreatic cancer (LAPC).
Methods
Among one-hundred fourteen patients with LAPC treated at our institution with initial chemotherapy followed by curative chemoradiation (CRT) from July 2013 to March 2022, a secondary analysis with baseline 18F-FDG PET/CT images was conducted in fifty-seven patients. All pre-treatment PET/CT were performed at a single PET/CT Centre. Clinical factors as well as semiquantitative PET parameters, including standardized uptake value (SUV), metabolic tumor volume (MTV), and total lesion glycolysis (TLG), were also reported. Early progression (EP) was defined temporally as a progression at the first evaluation, at 3 months from the start of treatment. EP was evaluated by CT scan, resulting in a dichotomous label of progression. A 3D Volume of Interest (VOI) was placed over the primary tumour, manually delineated. Three families of hand-crafted features were extracted from the VOIs of each patient's images, from both CT and PET acquisitions, thus quantifying grey intensity and tissue texture. Statistical features consisted of the moments up to the fourth-order of the first-order image histogram, i.e., the mean, the standard deviation, the skewness and the kurtosis. Texture features were derived from the 3D gray-level co-occurrence matrix (GLCM) and from the Local Binary Patterns-TOP (LBP-TOP). The final dataset was then created by adding clinical data from each patient. The predictive pipeline consisted of a feature selection phase followed by a sequence of two cascading decision trees in which the second used the predictions of the first as additional features for sample prediction. In the training phase, this model optimised the binarization threshold for classification to be applied later in the testing phase. The whole system follows a ten fold cross-validation approach. The quality of the proposed model was appraised by means of receiver operating characteristics (ROC) and areas under the ROC curve (AUC).
Results
Given each 3D VOI in the images, we computed the radiomics features, taking into consideration 12 statistical features, 230 textural features (182 GLCM, 48 LBP-TOP) extracted from the images, and adding 15 clinical features. We defined the final performance. To the best of our knowledge, this is the first study for feasibility and hypothesis generation of a radiomic strategy to predict early progression in LAPC and our data suggests that a specific signature can be identified (AUC 0.83; prediction accuracy 80.7%).
Conclusions
This model based on clinical and PET/CT radiomic features assessed before treatment can predict the early progression in LAPC patients. It could be a promising pre-treatment, non-invasive, approach that can assist physicians in evaluating the risk of early progression in patients individually, and thus achieving a personalized treatment and better clinical outcomes. The identification of the external validation dataset is actually ongoing.
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosures
All authors have declared no conflicts of interest.
