Jakoba J Eertink1,2, Gerben J C Zwezerijnen3,4, Matthijs C F Cysouw3,4, Sanne E Wiegers5,3, Elisabeth A G Pfaehler6, Pieternella J Lugtenburg7, Bronno van der Holt8, Otto S Hoekstra3,4, Henrica C W de Vet9,10, Josée M Zijlstra5,3, Ronald Boellaard3,4. 1. Department of Hematology, Amsterdam UMC Location Vrije Universiteit Amsterdam, De Boelelaan 1117, 1081 HV, Amsterdam, The Netherlands. j.eertink@amsterdamumc.nl. 2. Cancer Center Amsterdam, Imaging and Biomarkers, Amsterdam, The Netherlands. j.eertink@amsterdamumc.nl. 3. Cancer Center Amsterdam, Imaging and Biomarkers, Amsterdam, The Netherlands. 4. Radiology and Nuclear Medicine, Amsterdam UMC Location Vrije Universiteit Amsterdam, Amsterdam, The Netherlands. 5. Department of Hematology, Amsterdam UMC Location Vrije Universiteit Amsterdam, De Boelelaan 1117, 1081 HV, Amsterdam, The Netherlands. 6. Department of Nuclear Medicine, University Hospital Augsburg, Augsburg, Germany. 7. Department of Hematology, Erasmus MC Cancer Institute, University Medical Center Rotterdam, Wytemaweg 80, 3015 CN, Rotterdam, the Netherlands. 8. Department of Hematology, HOVON Data Center, Erasmus MC Cancer Institute, Dr. Molewaterplein 40, 3015 GD, Rotterdam, the Netherlands. 9. Epidemiology and Data Science, Amsterdam UMC Location Vrije Universiteit Amsterdam, Amsterdam, The Netherlands. 10. Amsterdam Public Health Research Institute, Methodology, Amsterdam, The Netherlands.
Abstract
PURPOSE: Biomarkers that can accurately predict outcome in DLBCL patients are urgently needed. Radiomics features extracted from baseline [18F]-FDG PET/CT scans have shown promising results. This study aims to investigate which lesion- and feature-selection approaches/methods resulted in the best prediction of progression after 2 years. METHODS: A total of 296 patients were included. 485 radiomics features (n = 5 conventional PET, n = 22 morphology, n = 50 intensity, n = 408 texture) were extracted for all individual lesions and at patient level, where all lesions were aggregated into one VOI. 18 features quantifying dissemination were extracted at patient level. Several lesion selection approaches were tested (largest or hottest lesion, patient level [all with/without dissemination], maximum or median of all lesions) and compared to the predictive value of our previously published model. Several data reduction methods were applied (principal component analysis, recursive feature elimination (RFE), factor analysis, and univariate selection). The predictive value of all models was tested using a fivefold cross-validation approach with 50 repeats with and without oversampling, yielding the mean cross-validated AUC (CV-AUC). Additionally, the relative importance of individual radiomics features was determined. RESULTS: Models with conventional PET and dissemination features showed the highest predictive value (CV-AUC: 0.72-0.75). Dissemination features had the highest relative importance in these models. No lesion selection approach showed significantly higher predictive value compared to our previous model. Oversampling combined with RFE resulted in highest CV-AUCs. CONCLUSION: Regardless of the applied lesion selection or feature selection approach and feature reduction methods, patient level conventional PET features and dissemination features have the highest predictive value. Trial registration number and date: EudraCT: 2006-005174-42, 01-08-2008.
PURPOSE: Biomarkers that can accurately predict outcome in DLBCL patients are urgently needed. Radiomics features extracted from baseline [18F]-FDG PET/CT scans have shown promising results. This study aims to investigate which lesion- and feature-selection approaches/methods resulted in the best prediction of progression after 2 years. METHODS: A total of 296 patients were included. 485 radiomics features (n = 5 conventional PET, n = 22 morphology, n = 50 intensity, n = 408 texture) were extracted for all individual lesions and at patient level, where all lesions were aggregated into one VOI. 18 features quantifying dissemination were extracted at patient level. Several lesion selection approaches were tested (largest or hottest lesion, patient level [all with/without dissemination], maximum or median of all lesions) and compared to the predictive value of our previously published model. Several data reduction methods were applied (principal component analysis, recursive feature elimination (RFE), factor analysis, and univariate selection). The predictive value of all models was tested using a fivefold cross-validation approach with 50 repeats with and without oversampling, yielding the mean cross-validated AUC (CV-AUC). Additionally, the relative importance of individual radiomics features was determined. RESULTS: Models with conventional PET and dissemination features showed the highest predictive value (CV-AUC: 0.72-0.75). Dissemination features had the highest relative importance in these models. No lesion selection approach showed significantly higher predictive value compared to our previous model. Oversampling combined with RFE resulted in highest CV-AUCs. CONCLUSION: Regardless of the applied lesion selection or feature selection approach and feature reduction methods, patient level conventional PET features and dissemination features have the highest predictive value. Trial registration number and date: EudraCT: 2006-005174-42, 01-08-2008.
Authors: Michael Crump; Sattva S Neelapu; Umar Farooq; Eric Van Den Neste; John Kuruvilla; Jason Westin; Brian K Link; Annette Hay; James R Cerhan; Liting Zhu; Sami Boussetta; Lei Feng; Matthew J Maurer; Lynn Navale; Jeff Wiezorek; William Y Go; Christian Gisselbrecht Journal: Blood Date: 2017-08-03 Impact factor: 22.113
Authors: Thomas M Habermann; Edie A Weller; Vicki A Morrison; Randy D Gascoyne; Peter A Cassileth; Jeffrey B Cohn; Shaker R Dakhil; Bruce Woda; Richard I Fisher; Bruce A Peterson; Sandra J Horning Journal: J Clin Oncol Date: 2006-06-05 Impact factor: 44.544