27 Radiomics of Chronic Lymphocytic Leukemia: multi-label whole body MRI segmentation

Introduction The arrival of new molecules for Chronic Lymphocytic Leukemia patient care involves a longitudinal follow-up to assess the response to treatment. The aim of this study is to segment the organs of interest in order to extract imaging biomarkers from MRI, CT and PET modalities. Methods Patients included in the trial undergo an imaging protocol of MRI and PET/CT scans at M0, before treatment, M1, M12 and M24 follow-up times. A segmentation based on Fast-Marching with regularization term was developed [1] and consists in dropping “organ” and “background” seeds to delineate MRI organs of interest, i.e. cervical, axillary, inguinal, iliac, retroperitoneal, mediastinal lymph nodes as well as liver, spleen and bone marrow. The process is interactive and iterative as the user can add or move the seeds to relaunch and sharpen the segmentation. Each organ (10) defined on the MR modality is characterized by 106 features (dimensions), extracted from the module Radiomics [2] of 3D Slicer. Results From MR images of 3 patients, mean segmentation times were 98, 132 and 56 min with a clinical contouring tool, the times decreased to 31,31 and 19 min with the implemented segmentation. Fig. 1 presents the results of the implemented multi-label whole body MRI segmentation at M0, M1 and M12 for the same patient. The feature groups are presented in Table 1. A distribution assessment of the Radiomics features by organ was considered. Fig. 2 shows the distributions of these organs in 2D after dimension reduction (106 dimensions in the Radiomics features space reduced to 2D) through the t-SNE algorithm. t-SNE algorithm allows us to optimally preserve the distance between observational points in the original space. One can notice i) the proximity of each organ through their features and ii) a similar patient distribution inside each group of organs. Conclusions The multi-label whole body MRI segmentation allows us to study the evolution of the organs of interest. In addition to the variation follow-up of simple parameters such as volume, other Radiomics features have been extracted. These Radiomics data will be combined with other data from biochemical samplings. A study of these data from heterogeneous nature and coming from different temporality will be also considered. Download high-res image (249KB) Download full-size image Download high-res image (158KB) Download full-size image Download high-res image (495KB) Download full-size image