Generation and validation of a PET radiomics model that predicts survival in diffuse large B cell lymphoma treated with R‐CHOP14: A SAKK 38/07 trial post‐hoc analysis

Abstract

Functional parameters from positron emission tomography (PET) seem promising biomarkers in various lymphoma subtypes. This study investigated the prognostic value of PET radiomics in diffuse large B‐cell lymphoma (DLBCL) patients treated with R‐CHOP given either every 14 (testing set) or 21 days (validation set). Using the PyRadiomics Python package, 107 radiomics features were extracted from baseline PET scans of 133 patients enrolled in the Swiss Group for Clinical Cancer Research 38/07 prospective clinical trial (SAKK 38/07) [ClinicalTrial.gov identifier: NCT00544219]. The international prognostic indices, the main clinical parameters and standard PET metrics, together with 52 radiomics uncorrelated features (selected using the Spearman correlation test) were included in a least absolute shrinkage and selection operator (LASSO) Cox regression to assess their impact on progression‐free (PFS), cause‐specific (CSS), and overall survival (OS). A linear combination of the resulting parameters generated a prognostic radiomics score (RS) whose area under the curve (AUC) was calculated by receiver operating characteristic analysis. The RS efficacy was validated in an independent cohort of 107 DLBCL patients. LASSO Cox regression identified four radiomics features predicting PFS in SAKK 38/07. The derived RS showed a significant capability to foresee PFS in both testing (AUC, 0.709; p < 0.001) and validation (AUC, 0.706; p < 0.001) sets. RS was significantly associated also with CSS and OS in testing (CSS: AUC, 0.721; p < 0.001; OS: AUC, 0.740; p < 0.001) and validation (CSS: AUC, 0.763; p < 0.0001; OS: AUC, 0.703; p = 0.004) sets. The RS allowed risk classification of patients with significantly different PFS, CSS, and OS in both cohorts showing better predictive accuracy respect to clinical international indices. PET‐derived radiomics may improve the prediction of outcome in DLBCL patients.

Radiomics in Malignant Lymphomas

“…They found a single heterogeneity-related radiomic feature (GLRLM run length non-uniformity, RLN), to be associated with PFS and OS (45) whilst another larger study of 132 DLBCL patients, confirmed the prognostic value of the MH of the largest tumor lesion (described by a radiomic feature termed Long-Zone High-Grey Level Emphasis, LZHGE), in predicting event-free survival (46). Finally, two most recent studies, one from the Swiss Group for Clinical Cancer Research (SAKK) (47) and the other from the Key Laboratory of Medical Molecular Imaging of Zhejiang Province, Hangzhou, China explored the radiomics in DLBCL using a similar approach (48). Both these studies applied the least absolute shrinkage and selection operator (LASSO) algorithm to select the features that define the radiomic signatures (RS), both built the RS in a testing set and validated their findings in a separate patient cohort.…”

“…The linear combination of the selected radiomic features generated a prognostic radiomics score whose prognostic efficacy was validated in an independent cohort of 107 DLBCL patients treated with the R-CHOP21 regimen. The radiomic signature (RS) allowed risk classification of patients with significantly different PFS, and OS in both cohorts showing better predictive accuracy respect to clinical international indices (47).…”

Radiomics in Oncological PET Imaging: A Systematic Review—Part 2, Infradiaphragmatic Cancers, Blood Malignancies, Melanoma and Musculoskeletal Cancers

“…A total of 24 articles on lymphomas were included in this review [ 122 , 123 , 124 , 125 , 126 , 127 , 128 , 129 , 130 , 131 , 132 , 133 , 134 , 135 , 136 , 137 , 138 , 139 , 140 , 141 , 142 , 143 , 144 , 145 ], 13 of which studying diffuse large B-cell lymphoma (including 2 studies on gastro-intestinal lymphoma), 3 on follicular lymphoma, 3 on Hodgkin’s lymphoma, 2 on mantle cell lymphoma and 3 on other sub-types of lymphoma. 18F-FDG was the only tracer employed and all studies built radiomic models on baseline, pre-treatment PET images, often including clinical parameters and international prognostic indices.…”

Discovery of Pre-Treatment FDG PET/CT-Derived Radiomics-Based Models for Predicting Outcome in Diffuse Large B-Cell Lymphoma

“…Other studies which have explored the use of radiomic features in outcome prediction in DLBCL are not always directly comparable [ 12 , 28 , 29 , 30 , 31 , 32 ]. This is mainly due to differences in segmentation methodology, modelling techniques and outcome measures between groups.…”

“…It may be that the same features would be chosen between the different datasets, but each method removes the alternate correlated feature and, therefore, appears to create an entirely new model. Both Zhang et al and Ceriani et al used lasso in their cox regression models to select the most appropriate features [ 31 , 32 ]. Zhang et al in a study of 152 patients (training = 100, validation = 52) treated with R-CHOP or R-EPOCH (rituximab, etoposide, prednisone, vincristine, cyclophosphamide, and doxorubicin) found that a survival model created with radiomic features and MTV had a validation time dependent ROC AUC of 0.748 (95% CI 0.596–0.886).…”