Lung nodule detection in 2D radiographs

Written by: Clémentine Hatton and Audrey Henry

Published on: Jun, 2025

Summary

Deep learning models are a promising approach to enhance medical image analysis. However, their performance is limited by the size, quality, and diversity of available training datasets. Incorporating synthetic data into training datasets can address the lack of medical imaging data. The combined use of a virtual anthropomorphic model with gVirtualXRay enables the rapid generation of large synthetic datasets. By injecting artificial lesions, deforming organ shapes and adjusting scanner settings, it is possible to replicate the variability found in real datasets. In the case of pulmonary nodule detection, training a YOLOv8 detection model on a mixed dataset composed of real and synthetic radiographs generated by gVXR, improved nodule detection performance by 32%, compared to training exclusively on real data. These preliminary results underscore the potential of gVXR to generate automatically labelled synthetic data of sufficient quality for training diagnostic models, thereby addressing the challenge of data scarcity in medical imaging research.

Award

3^rd place: Vidal, F. P., Afshari, S., Albiol, A., Albiol, F., Bellot, A., Brun, A. L., Chou, C.-Y., Desbarats, P., García, M., Giovannelli, J.-F., Hatton, C., Henry, A., Kelly, G., Michelet, C., Mihail, R. P., Racy, M., Rouwane, A., Seznec, H., Sújar, A., Tugwell-Allsup, J., and Villard, P.-F.: "X-ray simulations with gVirtualXray in medicine and life sciences". Dirk Bartz Prize for Visual Computing in Medicine and Life Sciences 2025 (Eurographics Medical Prize).

Publication

Vidal, F. P., Afshari, S., Albiol, A., Albiol, F., Bellot, A., Brun, A. L., Chou, C.-Y., Desbarats, P., García, M., Giovannelli, J.-F., Hatton, C., Henry, A., Kelly, G., Michelet, C., Mihail, R. P., Racy, M., Rouwane, A., Seznec, H., Sújar, A., Tugwell-Allsup, J., and Villard, P.-F. (2025). Meuschke, M. & Kuhlen, T. W. (Eds.). “X-ray simulations with gVirtualXray in medicine and life sciences”. Dirk Bartz Prize for Visual Computing in Medicine and Life Sciences 2025 (Eurographics Medical Prize). DOI: 10.2312/evm.20251974

Citation

@INPROCEEDINGS{Vidal2025DirkBartz,
  author={F. P. Vidal and S. Afshari and A. Albiol and F. Albiol and 
    A. Corb\'i Bellot and A. L. Brun and C.-Y. Chou and P. Desbarats and 
    M. Garc\'ia and J.-F. Giovannelli and     C. Hatton and A. Henry and 
    G. Kelly and C. Michelet and Radu P. Mihail M. Racy and A. Rouwane and
    H. Seznec and A. S\'ujar and J. Tugwell-Allsup and P.-F. Villard},
  booktitle={Dirk Bartz Prize for Visual Computing in Medicine and 
    Life Sciences 2025 (Eurographics Medical Prize)}, 
  title={{X-ray simulations with gVirtualXray in medicine and life sciences}}, 
  year=2025,
  editor = {Monique Meuschke and Torsten W. Kuhlen},
  publisher = {The Eurographics Association},
  ISBN = {978-3-03868-281-3},
  DOI = {10.2312/evm.20251974}
}