17th International Conference on Malignant Lymphoma, Palazzo dei Congressi, Lugano, Switzerland, 13 - 17 June, 2023.

Abstract

Introduction: Pre-treatment PET/CT scans have not yet been incorporated into risk stratification models despite being the imaging modality of choice in DLBCL. Deep learning computer algorithms have consistently demonstrated the ability to analyze imaging data and identify features that are not readily apparent to the human eye. Therefore, to assess the prognostic performance of pre-treatment PET/CT scans in patients with DLBCL, we compared automatically extracted radiomics features using a convolutional neural network with a standard prognosticating tool (NCCN-IPI). Methods: Native DICOM images from FDG PET/CT scans of newly diagnosed DLBCL patients were preprocessed for regularity and then segmented using the nnU-Net architecture with an external lymphoma-specific model. Feature extraction and SUV characterization were accomplished with PyRadiomics and NiBabel. Our computational method included no manual segmentation correction. Individual radiomics features (n = 115) were evaluated for association with event free survival (EFS) via univariate Cox regression analysis and by Kaplan Meier log rank testing of upper and lower feature quartiles. Additionally, Spearman and Pearson correlation matrices were used to guide feature selection while minimizing collinearity. We then assessed the association of the highest scoring radiomics features with EFS. Results: Radiomics features were automatically extracted from a cohort of 713 newly diagnosed DLBCL patients with pre-treatment PET/CT scans treated at Mayo Clinic between 2003 and 2015 and followed through 2023. Median age at diagnosis was 64 years (range 18–93), 41% were females. NCCN-IPI composition and outcomes of our cohort was comparable to the previously published NCCN cohort, with patient distribution and 5-year OS as follows: Low (11%, 5-y OS 89%), Low-intermediate (42%, 5-y OS 72%), High-intermediate (38%, 5-y OS 53%), High (9%, 5-y OS 25%). Log rank testing found numerous radiomics features were associated with EFS including least axis length (LAL) (HR: 2.83, 95% CI: 2.08–3.84, p < 0.005, Figure A) and volume (HR: 2.47, 95% CI: 1.84–3.01, p < 0.005, Figure B). As a reference, NCCN-IPI univariate Cox regression analysis of EFS produced a Harrell’s C-statistic (c) of 0.633. Single radiomics features of sphericity, volume, surface area, and least axis length individually produced unadjusted c-statistics of 0.624, 0.629, 0.636 and 0.638, respectively. Combinations of radiomics features (surface area + least axis length) demonstrated cumulative benefit (c = 0.642). Keywords: Aggressive B-cell non-Hodgkin lymphoma, PET-CT, Risk Models Conflicts of interests pertinent to the abstract. T. M Habermann Consultant or advisory role: Data Monitoring Committee: Seagen, Tess Therapeutics, Eli Lilly & Co. Scientific Advisory Board: Morpohsys, Incyte, Biegene, Loxo Oncology. No personal compensation is received for these activities, any compensation is received by my institution. Research funding: Genentech, Sorrento, BMS