Radiomics model of pelvic bone marrow in MRI for prediction of plasma cell infiltration in multiple myeloma patients

abstract

English

Introduction: Plasma cell infiltration (PCI) is an important factor for staging and risk stratification in patients with monoclonal plasma cell disorders but can only be obtained by invasive biopsy. Radiomics is a new method that allows for objective, in-depth tissue characterization by non-invasive imaging by calculating a large number of histogram, shape and textural features from medical images. The purpose of this work was to investigate with which accuracy radiomics models can predict PCI results from unguided biopsy at the iliac crest. Methods: One hundred fifty-eight patients with smoldering or multiple myeloma who had undergone whole body-MRI at 1.5 Tesla as well as bone marrow (BM) biopsy were included. Data was split by date of the MRI in training set (n=116) and independent test set (n=42). BM of the right and left hip bone was segmented in coronal T1 turbo-spin-echo images. Two hundred thirty-three radiomics features were calculated for each hipbone. A random forest classifier was trained to predict PCI from radiomics features on the training set and the model was evaluated in the independent test set. Mean absolute error (MAE) in [%PCI] reports the accuracy of the PCI prediction. For comparison, a linear model correlating the mean T1-BM-signal intensity of the pelvis (normalized to muscle) to the PCI was established on the training set and applied to the test set (mean intensity model). Additionally, two radiologists rated the diffuse infiltration (DI) according to 3 levels of severity (none-to-mild vs. moderate vs. severe). The mean PCI within each severity level from the training set was determined and assigned as a prediction to patients with the same level diffuse infiltration in the test set (the radiologists’ prediction). Results: The MAE of the mean intensity model was 23.4 [%PCI]. The MAE of the radiomics model was 14.3 [%PCI]. The MAE of the radiologists’ prediction of raters 1 and 2 were 16.1 [%PCI] and 16.7 [%PCI], respectively. Conclusions: We established a radiomics model to predict PCI from T1 MR images of the pelvic BM. The radiomics model, which also analyses textural features of the BM, performs markedly better than a model based on mean T1-BM signal intensity only and similar to the prediction by two radiologists based on DI severity. Further improvement of the model is necessary, which might be obtained by enlarging the amount of training data and adding T2 and diffusion-weighted imaging based radiomics features in the model.

multiple myeloma

UHLENBROCK, C.; WENNMANN, M.; KLEIN, A.; BAUER, F.; CHMELÍK, J.; GRÖZINGER, M.; ROTKOPF, L.; GOTZ, M.; THIERJUNG, H.; SAUER, S.; BONEKAMP, D.; KLEESIEK, J.; WEBER, T.; HILLENGASS, J.; SCHLEMMER, H.; GOLDSCHMIDT, H.; FLOCA, R.; WEINHOLD, N.; MAIER-HEIN, K.; DELORME, S.

28. 9. 2021

KARGER, ALLSCHWILERSTRASSE 10, CH-4009 BASEL, SWITZERLAND

978-3-318-06712-5

Jahrestagung der Deutschen, Österreichischen und Schweizerischen Gesellschaften für Hämatologie und Medizinische Onkologie

2296-5270

Oncology Research and Treatment

44

S2

unknown

84

84

1

https://www.karger.com/Article/Pdf/518417