Advanced Automated Model for Robust Bone Marrow Segmentation in Whole-body MRI

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Monoclonal plasma cell disorders (MPCD) require accurate assessment of bone marrow involvement.
• Whole-body MRI (WB-MRI) is a key imaging modality for MPCD.
• Automated segmentation of bone marrow spaces (BMS) can improve efficiency and consistency.

Purpose of the Study:

• To develop an advanced automated BMS segmentation model using nnU-Net for WB-MRI in MPCD.
• To validate the model's performance on multicenter datasets with diverse pathologies.
• To compare the advanced model against a previously established basic segmentation model.

Main Methods:

• Utilized a cohort of 210 WB-MRIs from 207 MPCD patients across 8 centers.
• Trained an nnU-Net algorithm on 186 T1-weighted WB-MRIs.
• Tested the model on independent datasets, including cases with extensive tumor load and varying image quality.

Main Results:

• Achieved high mean Dice scores for BMS segmentation: 0.89±0.13 (test set I) and 0.88±0.11 (test set II).
• Demonstrated significantly improved performance compared to a prior basic model (p<0.05).
• Accurately segmented bone marrow affected by pathologies, artifacts, and low imaging quality.

Conclusions:

• The advanced automated model provides reliable segmentation of BMS on multicenter WB-MRI data.
• The model is robust even in the presence of severe myeloma-related pathologies and imaging heterogeneity.
• This technology holds promise for improved quantitative analysis in MPCD management.