Deep Learning Model Improves Abdominal Organ Segmentation Accuracy

Strategic Insights -

Researchers at Cedars-Sinai Medical Center have developed a deep learning segmentation model that can accurately segment abdominal organs on CT and MRI scans. The model was trained using a domain randomization method, which improves its generalization ability on cross-site and cross-modality datasets. The model outperformed two other publicly available segmentation models on data from unseen test domains, achieving a mean Dice similarity coefficient of 0.88, compared to 0.79.

Key Takeaways:

  • The researchers developed a deep learning segmentation model that can segment abdominal organs on CT and MRI scans with high accuracy and generalization ability.
  • The model was trained using an extended nnU-Net model and a domain randomization method that improves its ability to generalize across different datasets and modalities.
  • The domain randomization method was used to train the model on the abdominal multiorgan segmentation challenge dataset, and it was compared to two commonly used segmentation algorithms (TotalSegmentator and MRSegmentator).
  • The proposed domain randomization method showed improved generalization ability on the cross-site and cross-modality datasets compared to state-of-the-art methods.
  • The segmentation model using the domain randomization method outperformed two other publicly available segmentation models on data from unseen test domains, achieving a mean Dice similarity coefficient of 0.88.
  • The model was evaluated using the Dice similarity coefficient (DSC) and was found to outperform two other segmentation algorithms on data from unseen test domains.
  • The research was conducted by Lixia Wang, Yu Shi, Touseef Ahmad Qureshi, Zengtian Deng, Yibin Xie, and Debiao Li from the Biomedical Imaging Research Institute at Cedars-Sinai Medical Center.

Statistics:

  • 0.88: mean Dice similarity coefficient of the segmentation model using the domain randomization method
  • 0.79: mean Dice similarity coefficient of two other publicly available segmentation models
  • 0.88 vs 0.79: comparison of the mean Dice similarity coefficient of the segmentation model using the domain randomization method and two other publicly available segmentation models
  • 2025: publication year of the research
  • Radiology: 7(4): journal volume and issue number where the research was published
  • Hindawi Publishing: publisher of the research
  • RSNA: publisher of the research
  • $conference_name: name of the conference or organization that published the research

Sources:

  • Deep Learning with Domain Randomization in Image and Feature Spaces for Abdominal Multiorgan Segmentation on CT and MRI Scans. Radiology, 2025;7(4).
  • Hindawi Publishing. www.hindawi.com.
  • Radiology. www.hindawi.com/journals/rrp/.
  • NewsRx. Investigators at Cedars-Sinai Medical Center Target Artificial Intelligence (Deep Learning with Domain Randomization in Image and Feature Spaces for Abdominal Multiorgan Segmentation on CT and MRI Scans). Robotics & Machine Learning. October 13, 2025; p 239.