Hasso-Plattner-Institut
Prof. Dr. Christoph Lippert
 

Medical Imaging - Reducing the dependency on expert supervision

CLASSIFAI - Usable Active Learning for medical image segmentation

CLASSIFAI is a user-centered learning system for interactive annotation and segmentation of medical image data for therapy planning and execution of scientific studies - an Artificial Intelligence with the human being in the center. Classifai makes active machine learning useful for the analysis of volumetric images in medicine, e.g. MRIs. This approach has the potential to massively accelerate extensive components of the work of medical researchers and physicians. It also opens up the possibility of accelerating the performance of studies, diagnoses and therapies.

https://www.youtube.com/embed/xqpGiga8Scw

Self-Supervised Learning for Medical Imaging

Self-supervised learning methods have witnessed a recent surge of interest after proving successful in multiple application fields. We propose 3D versions for five different self-supervised methods, in the form of proxy tasks. Our methods facilitate neural network feature learning from unlabeled 3D images, aiming to reduce the required cost for expert annotation. The developed algorithms are 3D Contrastive Predictive Coding, 3D Rotation prediction, 3D Jigsaw puzzles, Relative 3D patch location, and 3D Exemplar networks.

https://github.com/HealthML/self-supervised-3d-tasks

Taleb, A., Loetzsch, W., Danz, N., Severin, J., Gaertner, T., Bergner, B., & Lippert, C. (2020). 3D Self-Supervised Methods for Medical Imaging. Neural Information Processing Systems (NeurIPS) arXiv:2006.03829.

Taleb, A., Lippert, C., Klein, T., & Nabi, M. (2019). Multimodal self-supervised learning for medical image analysis. arXiv preprint arXiv:1912.05396.

Generative Adversarial Networks for Medical Imaging

Imbalanced training data introduce important challenge into medical image analysis where a majority of the data belongs to a normal class and only few samples belong to abnormal classes. We propose to mitigate the class imbalance problem by introducing two generative adversarial network (GAN) architectures for class minority oversampling.

Rezaei, M., Näppi, J. J., Lippert, C., Meinel, C., & Yoshida, H. (2020). Generative multi-adversarial network for striking the right balance in abdominal image segmentation. International Journal of Computer Assisted Radiology and Surgery, 1-12.

Rezaei, M., Uemura, T., Näppi, J., Yoshida, H., Lippert, C., & Meinel, C. (2020). Generative synthetic adversarial network for internal bias correction and handling class imbalance problem in medical image diagnosis. In Medical Imaging 2020: Computer-Aided Diagnosis (Vol. 11314, p. 113140E). International Society for Optics and Photonics.