Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Veriserum: A dual-plane fluoroscopic dataset with knee implant phantoms for deep learning in medical imaging

Jinhao Wang, Florian Vogl, Pascal Schütz, Saša Ćuković, William R. Taylor

TL;DR

Veriserum tackles the shortage of open, realistic dual-plane fluoroscopic knee data by providing ~110k images with ground-truth poses derived from a precision robot and calibrated via DISCAL/SICAL, enabling reliable benchmarking of 2D/3D registration. The dataset supports tasks such as 2D/3D registration, segmentation, and reconstruction, and includes a differentiable renderer-based automated pose registration that closely matches manual ground truth. Automated poses achieve submillimeter translational and sub-degree rotational accuracy on most samples, offering a scalable, reproducible alternative to time-consuming manual annotation. Despite lacking soft-tissue anatomy and using legacy imaging hardware, Veriserum offers a valuable stepping-stone toward AI-assisted knee imaging pipelines and tighter integration of traditional and data-driven methods.

Abstract

Veriserum is an open-source dataset designed to support the training of deep learning registration for dual-plane fluoroscopic analysis. It comprises approximately 110,000 X-ray images of 10 knee implant pair combinations (2 femur and 5 tibia implants) captured during 1,600 trials, incorporating poses associated with daily activities such as level gait and ramp descent. Each image is annotated with an automatically registered ground-truth pose, while 200 images include manually registered poses for benchmarking. Key features of Veriserum include dual-plane images and calibration tools. The dataset aims to support the development of applications such as 2D/3D image registration, image segmentation, X-ray distortion correction, and 3D reconstruction. Freely accessible, Veriserum aims to advance computer vision and medical imaging research by providing a reproducible benchmark for algorithm development and evaluation. The Veriserum dataset used in this study is publicly available via https://movement.ethz.ch/data-repository/veriserum.html, with the data stored at ETH Zürich Research Collections: https://doi.org/10.3929/ethz-b-000701146.

Veriserum: A dual-plane fluoroscopic dataset with knee implant phantoms for deep learning in medical imaging

TL;DR

Veriserum tackles the shortage of open, realistic dual-plane fluoroscopic knee data by providing ~110k images with ground-truth poses derived from a precision robot and calibrated via DISCAL/SICAL, enabling reliable benchmarking of 2D/3D registration. The dataset supports tasks such as 2D/3D registration, segmentation, and reconstruction, and includes a differentiable renderer-based automated pose registration that closely matches manual ground truth. Automated poses achieve submillimeter translational and sub-degree rotational accuracy on most samples, offering a scalable, reproducible alternative to time-consuming manual annotation. Despite lacking soft-tissue anatomy and using legacy imaging hardware, Veriserum offers a valuable stepping-stone toward AI-assisted knee imaging pipelines and tighter integration of traditional and data-driven methods.

Abstract

Veriserum is an open-source dataset designed to support the training of deep learning registration for dual-plane fluoroscopic analysis. It comprises approximately 110,000 X-ray images of 10 knee implant pair combinations (2 femur and 5 tibia implants) captured during 1,600 trials, incorporating poses associated with daily activities such as level gait and ramp descent. Each image is annotated with an automatically registered ground-truth pose, while 200 images include manually registered poses for benchmarking. Key features of Veriserum include dual-plane images and calibration tools. The dataset aims to support the development of applications such as 2D/3D image registration, image segmentation, X-ray distortion correction, and 3D reconstruction. Freely accessible, Veriserum aims to advance computer vision and medical imaging research by providing a reproducible benchmark for algorithm development and evaluation. The Veriserum dataset used in this study is publicly available via https://movement.ethz.ch/data-repository/veriserum.html, with the data stored at ETH Zürich Research Collections: https://doi.org/10.3929/ethz-b-000701146.

Paper Structure

This paper contains 12 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Method
  3. Dual-plane fluoroscope and X-ray system
  4. Data acquisition and implant design
  5. Calibration measurements
  6. Poses
  7. Experiments and Results
  8. Data Collection and Repeatability
  9. Automated Pose Registration
  10. Conclusion and Discussion
  11. Acknowledgments.
  12. Disclosure of Interests.

Figures (5)

  • Figure 1: Schematic of the dual-plane X-ray imaging setup.
  • Figure 2: Calibration objects used in the dual-plane X-ray setup.
  • Figure 3: Comparison of different tibial and femoral implant designs: (a) Asymmetric round wings with fixed bearing, (b) Asymmetric straight wings with fixed bearing, (c) Symmetric round wings with fixed bearing, (d) Symmetric round wings with mobile bearing, (e) Femur crossbar design, (f) Femur normal design, (g) Asymmetric no-wings with fixed bearing.
  • Figure 4: Calibration images used in the experimental setup.
  • Figure 5: Error percentile plot of target robot poses and automatically registered poses (Auto-reg), compared to manual ground truth.