Generating fictive trigonocephaly data using a generative adversarial network to produce data to train deep learning algorithms
EANS Academy. Delye H. 09/27/19; 276001; EP12058
Dr. Hans Delye
Dr. Hans Delye

Access to this content is reserved for EANS members and attendees of this event. Click here to become an EANS member and gain your access to the full content of the EANS Academy

Discussion Forum (0)
Rate & Comment (0)
Introduction & Objectives: The early diagnosis of craniosynostosis is key to offer minimal invasive treatment options such as endoscopically assisted suturectomy. To aid in early diagnosis deep learning techniques such as neural networks could be used. Nevertheless, this requires a huge amount of data. Therefore, the use of another deep learning technique (a generative adversarial network, GAN) can help in generating fictive data similar to the real data. For this research a GAN was constructed to see if it could generate unique fictive heads suitable for neural network training.
Material & Methods: We used 3D data based on preoperative 3D stereophotogrammetry imaging of children with trigonocephaly. Several GANs were trained and qualified by comparing the sizes of the generated heads to the average real head. This resulted in the best version of the network. Next, the data was qualified by having professionals classify the generated heads as real or fake and by testing the generated heads in a neural network to see if the generated heads would be treated differently from the real heads.
Results: Due to a big difference in means and standard deviations only one version of the GAN was used for further qualification. All three tests performed, showed promising results for this network. The sizes of the generated heads were comparable to the sizes of the real heads (a difference of 0.71mm in the mean size and 0,22mm in the standard deviation of this mean), the professionals had difficulty to differentiate between real and fake heads and so had the neural network. The neural network used for testing showed the same accuracy (100%) and loss for both the generated and real heads.
Conclusion: Based on the described qualification tests, the fictive data generated by a GAN could be used to train neural networks to diagnose trigonocephaly
Code of conduct/disclaimer available in General Terms & Conditions
Anonymous User Privacy Preferences

Strictly Necessary Cookies (Always Active)

MULTILEARNING platforms and tools hereinafter referred as “MLG SOFTWARE” are provided to you as pure educational platforms/services requiring cookies to operate. In the case of the MLG SOFTWARE, cookies are essential for the Platform to function properly for the provision of education. If these cookies are disabled, a large subset of the functionality provided by the Platform will either be unavailable or cease to work as expected. The MLG SOFTWARE do not capture non-essential activities such as menu items and listings you click on or pages viewed.

Performance Cookies

Performance cookies are used to analyse how visitors use a website in order to provide a better user experience.

Google Analytics is used for user behavior tracking/reporting. Google Analytics works in parallel and independently from MLG’s features. Google Analytics relies on cookies and these cookies can be used by Google to track users across different platforms/services.

Save Settings