TotalSegmentator: robust segmentation of 104 anatomical structures in CT images (2208.05868v2)

Abstract: We present a deep learning segmentation model that can automatically and robustly segment all major anatomical structures in body CT images. In this retrospective study, 1204 CT examinations (from the years 2012, 2016, and 2020) were used to segment 104 anatomical structures (27 organs, 59 bones, 10 muscles, 8 vessels) relevant for use cases such as organ volumetry, disease characterization, and surgical or radiotherapy planning. The CT images were randomly sampled from routine clinical studies and thus represent a real-world dataset (different ages, pathologies, scanners, body parts, sequences, and sites). The authors trained an nnU-Net segmentation algorithm on this dataset and calculated Dice similarity coefficients (Dice) to evaluate the model's performance. The trained algorithm was applied to a second dataset of 4004 whole-body CT examinations to investigate age dependent volume and attenuation changes. The proposed model showed a high Dice score (0.943) on the test set, which included a wide range of clinical data with major pathologies. The model significantly outperformed another publicly available segmentation model on a separate dataset (Dice score, 0.932 versus 0.871, respectively). The aging study demonstrated significant correlations between age and volume and mean attenuation for a variety of organ groups (e.g., age and aortic volume; age and mean attenuation of the autochthonous dorsal musculature). The developed model enables robust and accurate segmentation of 104 anatomical structures. The annotated dataset (https://doi.org/10.5281/zenodo.6802613) and toolkit (https://www.github.com/wasserth/TotalSegmentator) are publicly available.

• The paper introduces TotalSegmentator, a deep learning model that segments 104 anatomical structures in CT images using the nnU-Net framework.
• It achieves a Dice score of 0.943, significantly outperforming a pretrained model and demonstrating robustness across diverse clinical conditions.
• The publicly available toolkit streamlines radiology workflows, supporting applications in volumetry, disease characterization, and surgical planning.

Overview of "TotalSegmentator: Robust Segmentation of 104 Anatomical Structures in CT Images"

The paper presents TotalSegmentator, a deep learning model designed for the comprehensive segmentation of 104 anatomical structures in CT images. Utilizing a robust dataset reflecting real-world clinical scenarios, the paper aims to streamline radiology workflows and enhance the precision in clinical applications.

Methodology

The research employs a retrospective dataset comprising 1204 CT scans from diverse years and settings. This dataset captures 104 anatomical structures, including organs, bones, muscles, and vessels, essential for applications such as volumetry, disease characterization, and surgical planning. The nnU-Net framework, chosen for its capability to automatically configure hyperparameters, served as the backbone model, allowing for efficient high-performance segmentation.

A detailed annotation process was executed, incorporating manual refinement and iterative learning to maximize accuracy. This process included utilizing existing models when available, followed by validation and refinement by experienced radiologists. Moreover, the paper compared TotalSegmentator to existing models using Dice similarity coefficients, demonstrating superior performance on an independent test set.

Results

The model achieved a Dice score of 0.943, outperforming a pretrained model with a score of 0.871 (p<0.001). Importantly, the model maintains robust performance across variable pathological conditions, demonstrating adaptability to real-world clinical complexities. Additionally, on a secondary dataset comprising 4004 whole-body CT scans, age-dependent alterations in volumetry and attenuation were explored. Notable correlations, such as between age and aortic volume (r = 0.64; p<0.001), were identified.

Implications and Future Directions

This work provides a publicly accessible, ready-to-use toolkit that significantly lowers the entry barrier for segmentation in medical imaging research. It facilitates quicker and more consistent segmentation across a variety of anatomical structures without requiring extensive radiological expertise or computational resources. Practically, this can accelerate diverse studies in radiology, potentially impacting surgical planning and timely diagnosis.

Theoretically, it sets a precedent for multi-organ segmentation models, pushing the field toward more holistic approaches in quantifying and analyzing anatomical features. The potential for future developments lies in extending the model's capabilities, such as incorporating more advanced architectures like transformers, further refining the segmentation accuracy and efficiency.

Conclusion

TotalSegmentator marks a significant advancement in anatomical segmentation. By providing a comprehensive and accessible tool with robust performance across real-world datasets, it not only meets current clinical needs but also paves the way for future innovations in medical imaging and diagnostics. This model can serve as a foundational component in expanding AI applications in radiology, with ongoing research anticipated to further enhance its scope and functionality.