Radiomics, pathomics and deep learning for precision medicine

Radiomics, pathomics and deep learning for precision medicine


The International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2017) is one the world’s premier forums for medical image analysis, and this year it has been organized in Quebec City, Canada. We are attending this exciting event, that started on Sunday 10th September, together with more than 1350 expected visitors.

Among the interesting topics introduced in the conference, I would like to focus on precision medicine presented in the engaging talk about “Radiomics, Pathomics and Deep Learning: Implications for Precision Medicine” given by Professor Anant Madabhushi yesterday. These are the main messages from the talk:

  • There is a strong need for image-based tests as they are comparatively cheap, can be deployed easily and yield complementary information to other, non-imaging-based testing methods, such as gene expression tests.
  • Understanding the decisions of deep neural networks is of critical importance in order to understand the – possibly sub-visual – features used by the network for making a decision, to make them more robust against subtle variations in the image acquisition process, and finally to approve developed algorithms from a regulatory point of view.
  • Besides taking the tumour heterogeneity into account, looking beyond the tumour will be more and more important, as not only the pathological tissue itself might reveal valuable information, but also its surrounding.
  • Efforts on integrating the temporal dimension are just at their very beginning. As each measurement only reflects a certain point in time of a long and complex disease development process, current achievements for fusing various spatial scales are good, but we have to go beyond that.

These are only some of the key takeaways from the first day, and we are looking forward to hearing more about the concrete steps into research in these directions during the upcoming days.