Workshop Aims

Image reconstruction is currently undergoing a paradigm shift that is driven by advances in machine learning. Whereas traditionally transform-based or optimization-based methods have dominated methods for image reconstruction, machine learning has opened up the opportunity for new data-driven approaches which have demonstrated a number of advantages over traditional approaches. In particular, deep learning techniques have shown significant potential for image reconstruction and offer interesting new approaches. Finally, machine learning approaches also offer the possibility for application-specific image reconstruction, e.g. in motion-compensated cardiac or fetal imaging.

This is supported by successes of machine learning in other inverse problems by multiple groups worldwide, with encouraging results and increasing interest. In the last year, we were able to attract more than 30 submissions for our workshop and could publish our own proceedings with 25 papers in LNCS with Springer. Hence, the topic has demonstrated to be of significant interest to the community. As in many other fields, machine learning is also driving progress in medical image reconstruction at a fast pace. In this respect, it is a fresh new way to push the boundaries of reconstruction algorithms by using extensive prior knowledge for superior diagnostic performance.

The aim of the workshop is to drive scientific discussion of advanced machine learning techniques for image acquisition and image reconstruction, opportunities for new applications as well as challenges in the evaluation and validation of ML based reconstruction approaches. Specifically, this will include topics such as those listed below (but not limited to):


  • Compressed sensing methods

  • Sparsity and low-rank methods

  • Machine learning for image super-resolution

  • Machine learning for image synthesis

  • Machine learning for quantitative imaging (including MRF)

  • Deep learning for image reconstruction including

    • CNN-based approaches

    • RNN-based approaches

    • Adversarial and generative approaches

  • Machine learning for

    • X-ray CT image reconstruction

    • MR image reconstruction

    • SPECT and PET image reconstruction

    • Ultrasound and optical imaging

    • Multimodality fusion or joint image reconstruction across two or more modalities

  • Applications of ML for image reconstruction in

    • Neuroimaging

    • Cardiac imaging

    • Abdominal imaging

    • Fetal and/or neonatal imaging

  • Validation of ML for image reconstruction