Publication detail

Deep Learning for Automatic Bone Marrow Apparent Diffusion Coefficient Measurements From Whole-Body Magnetic Resonance Imaging in Patients With Multiple Myeloma: A Retrospective Multicenter Study

WENNMANN, M. NEHER, P. STANZCYK, N. KIM-CELINE, K. KACHELE, J. WERU, V. HIELSCHER, T. GRÖZINGER, M. CHMELÍK, J. ZHANG, K. BAUER, F. NONNENMACHER, T. DEBIC, M. SAUER, S. ROTKOPF, L. JAUCH, A. SCHLAMP, K. MAI, E. WEINHOLD, N. AFAT, S. HORGER, M. GOLDSCHMIDT, H. SCHLEMMER, H. WEBER, T. DELORME, S. KURZ, F. MAIER-HEIN, K.

Original Title

Deep Learning for Automatic Bone Marrow Apparent Diffusion Coefficient Measurements From Whole-Body Magnetic Resonance Imaging in Patients With Multiple Myeloma: A Retrospective Multicenter Study

Type

journal article in Web of Science

Language

English

Original Abstract

Objectives: Diffusion-weighted magnetic resonance imaging plays an increasing role in patients with multiple myeloma. The objective of this study was to develop and test an algorithm for automatic pelvic bone marrow analysis from whole-body apparent diffusion coefficient maps in patients with multiple myeloma, by automatically segmentation of pelvic bones and subsequent extraction of objective, representative ADC measurements from each bone. Material and Methods: This retrospective multicentric study used 180 MRIs from 54 patients for developing an nnU-Net for automatic, individual segmentation of the right hip bone, the left hip bone, and the sacral bone. Precision of the automatic segmentation was tested on 15 wb-MRIs from 3 centers using the dice score. In three independent test-sets from three centers, which comprised a total of 312 whole-body MRIs, agreement between automatically extracted mean ADC values from the nnU-Net segmentation were compared to manual ADC-measurements by two radiologists. Bland-Altman plots were constructed, and absolute bias, relative bias to mean, limits of agreement, and coefficients of variation were calculated. In 56 patients with newly diagnosed multiple myeloma who had undergone bone marrow biopsy, ADC-values were correlated with biopsy results using Spearman correlation. Results: The ADC-nnU-Net achieved automatic segmentations with mean dice scores of 0.92, 0.93, and 0.85 for the right pelvis, the left pelvis, and the sacral bone, while the interrater experiment gave mean dice scores of 0.86, 0.86 and 0.77, respectively. The agreement between radiologists’ manual ADC measurements and automatic ADC measurements was as follows: the bias between the first rater and the automatic approach was 49 x10-6 mm2/s, 7 x10-6 mm2/s and -58 x10-6 mm2/s, and the bias between the second rater and the automatic approach was 12 x10-6 mm2/s, 2 x10-6 mm2/s and -66 x10-6 mm2/s for the right pelvis, the left pelvis, and the sacral bone. The bias between rater 1 and rater 2 was 40 x10-6 mm2/s, 8 x10-6 mm2/s and 7 x10-6 mm2/s, and the mean absolute difference between manual raters was 85 x10-6mm2/s, 65 x10-6mm2/s and 74 x10-6mm2/s. Automatically extracted ADC values significantly correlated with bone marrow plasma cell infiltration (R=0.36, p=0.007). Conclusion: In this study, a nnU-Net was developed which can automatically segment pelvic bone marrow from whole-body ADC-maps in multicentric data sets with a precision comparable to manual segmentations. This approach allows automatic, objective bone marrow ADC measurements which agree well with manual ADC measurements and can help to overcome interrater-variability or non-representative measurements. Automatically extracted ADC-values significantly correlate with bone marrow plasma cell infiltration and might therefore be of value for automatic staging, risk stratification, or therapy response assessment.

Keywords

deep learning; segmentation; diffusion-weighted imaging; apparent diffusion coefficient; whole-body; magnetic resonance imaging; multiple myeloma; monoclonal plasma cell disorders; bone marrow; multicentric

Authors

WENNMANN, M.; NEHER, P.; STANZCYK, N.; KIM-CELINE, K.; KACHELE, J.; WERU, V.; HIELSCHER, T.; GRÖZINGER, M.; CHMELÍK, J.; ZHANG, K.; BAUER, F.; NONNENMACHER, T.; DEBIC, M.; SAUER, S.; ROTKOPF, L.; JAUCH, A.; SCHLAMP, K.; MAI, E.; WEINHOLD, N.; AFAT, S.; HORGER, M.; GOLDSCHMIDT, H.; SCHLEMMER, H.; WEBER, T.; DELORME, S.; KURZ, F.; MAIER-HEIN, K.

Released

1. 4. 2023

Publisher

Wolters Kluwer Health, Inc.

ISBN

0020-9996

Periodical

INVESTIGATIVE RADIOLOGY

Year of study

58

Number

4

State

United States of America

Pages from

273

Pages to

282

Pages count

10

URL

BibTex

@article{BUT178933,
  author="Markus {Wennmann} and Peter {Neher} and Nikolas {Stanzcyk} and Kahl {Kim-Celine} and Jessica {Kachele} and Vivienn {Weru} and Thomas {Hielscher} and Martin {Grözinger} and Jiří {Chmelík} and Kevin Sun {Zhang} and Fabian {Bauer} and Tobias {Nonnenmacher} and Manuel {Debic} and Sandra {Sauer} and Lukas {Rotkopf} and Anna {Jauch} and Kai {Schlamp} and Elias {Mai} and Niels {Weinhold} and Saif {Afat} and Marius {Horger} and Hartmut {Goldschmidt} and Heinz-Peter {Schlemmer} and Tim {Weber} and Stefan {Delorme} and Felix {Kurz} and Klaus {Maier-Hein}",
  title="Deep Learning for Automatic Bone Marrow Apparent Diffusion Coefficient Measurements From Whole-Body Magnetic Resonance Imaging in Patients With Multiple Myeloma: A Retrospective Multicenter Study",
  journal="INVESTIGATIVE RADIOLOGY",
  year="2023",
  volume="58",
  number="4",
  pages="273--282",
  doi="10.1097/RLI.0000000000000932",
  issn="0020-9996",
  url="https://journals.lww.com/investigativeradiology/Abstract/9900/Deep_Learning_for_Automatic_Bone_Marrow_Apparent.64.aspx"
}