A machine learning based delta-radiomics process for early prediction of treatment response of pancreatic cancer

Abstract

Changes of radiomic features over time in longitudinal images, delta radiomics, can potentially be used as a biomarker to predict treatment response. This study aims to develop a delta-radiomic process based on machine learning by (1) acquiring and registering longitudinal images, (2) segmenting and populating regions of interest (ROIs), (3) extracting radiomic features and calculating their changes (delta-radiomic features, DRFs), (4) reducing feature space and determining candidate DRFs showing treatment-induced changes, and (5) creating outcome prediction models using machine learning. This process was demonstrated by retrospectively analyzing daily non-contrast CTs acquired during routine CT-guided-chemoradiation therapy for 90 pancreatic cancer patients. A total of 2520 CT sets (28-daily-fractions-per-patient) along with their pathological response were analyzed. Over 1300 radiomic features were extracted from the segmented ROIs. Highly correlated DRFs were ruled out using Spearman correlations. Correlation between the selected DRFs and pathological response was established using linear-regression-models. T test and linear-mixed-effects-models were used to determine which DRFs changed significantly compared with first fraction. A Bayesian-regularization-neural-network was used to build a response prediction model. The model was trained using 50 patients and leave-one-out-cross-validation. Performance was judged using the area-under-ROC-curve. External independent validation was done using data from the remaining 40 patients. The results show that 13 DRFs passed the tests and demonstrated significant changes following 2–4 weeks of treatment. The best performing combination differentiating good versus bad responders (CV-AUC = 0.94) was obtained using normalized-entropy-to-standard-deviation-difference-(NESTD), kurtosis, and coarseness. With further studies using larger data sets, delta radiomics may develop into a biomarker for early prediction of treatment response.

Improving Treatment Response Prediction for Chemoradiation Therapy of Pancreatic Cancer Using a Combination of Delta-Radiomics and the Clinical Biomarker CA19-9

“…Radiomics analysis has been associated with several clinical end points, several researchers showed that it has a potential to be developed into an imaging biomarker using different imaging modalities. We showed earlier that CT based radiomics has a potential in a variety of applications (11)(12)(13)(14)(15)(16)(17)(18)(19). Researchers also showed that it has promise using MR images for instance, Wibmer et al (33) and Vignti et al (34) showed the radiomics features as potential imaging biomarkers in prostate cancer.…”

“…Also, to reduce directional dependence of texture features, directionspecific matrices are summed and averaged to create a final global matrix. This resulted in 73 DRFs that could be used for deltaradiomic analysis (17)(18)(19). In this study, patients were divided into good-and poor-response groups based on their treatment responses.…”

“…In this study, patients were divided into good-and poor-response groups based on their treatment responses. Changes in each of the 73 DRFs were calculated as the relative net change of the radiomic feature of interest from the first fraction as follows (19):…”

“…A metric trend was established using a linear regression model to find the best fit for each of the DRFs features vs. response and subsequently determine those with potential trends. Features were also evaluated to determine if they changed during the treatment using a t-test and were tested for whether related to response by fitting linear mixed effects models using R R software for the change of DRFs as a function of response with two random effects to account for patient-and fraction-dependent variations as follows (19): The p-value of the log likelihood ratio for each model was calculated. The features that passed both the linear mixed effect model and the t-test (p < 0.05) were selected to be used for subsequent analysis.…”

“…Fave et al used DRFs to create a model for survival and distant metastases in lung and found that adding radiomic features improved the prognostic power of their model (16). Our previous analysis showed that DRFs including kurtosis, NESTD, and coarseness can predict treatment response for resectable pancreatic cancer with AUC of 0.94 (17)(18)(19). Although DRFs have been associated with several clinical endpoints in a variety of applications, the complex relationships of radiomics and clinical factors are largely unknown.…”