Publish on 29/01/2020
The remarkable achievements of AI in the health sector, to give just one example, have made it a promising new field for GAFA and BATX (their Chinese competitors). Caught up in the myriad media releases and fabulous promises, there is a growing need for clarity and education. Within this context, we met with Erwan Kerrien, an IT researcher in the Magrit team, a joint undertaking of the Inria Nancy - Grand Est centre and Loria. Erwan’s work is centred around the quest for artificial intelligent images.
Erwan Kerrien
© Inria / Photo J.F. Badias

Erwan Kerrien : " Inria has been working for a long time with Nancy CHRU (Regional University Hospital) and GE Healthcare (formerly General Electric Medical Systems). Doctors are good at taking a long-term view of their discipline, but they are not necessarily familiar with the technological options available to enable gradual improvements. It’s important to accompany doctors, to attend appointments and to experience difficulties as they are encountered. This is precisely where we come in, sharing our expertise in order to progress."

Where does AI fit into all this? It seems more like a collaboration involving human intelligence...

E.K  : "There is a tendency to reduce AI to Machine Learning and then to its sub-category Deep Learning. I am interested in images and being able to automatically understand what they contain. Images are three-dimensional worlds containing objects which, in this instance, for medical staff, will be a patient’s anatomy. My goal is to be able to develop readable representations of these three-dimensional worlds using real images. Here, the human eye and human intelligence are considerably more effective: doctors know how to read an image, distinguishing between bones and organs in order to make a diagnosis. Computers, meanwhile, only see points in space, pixels. How can we get from these points to something that is close to the understanding of a doctor?

I’m also very interested in real-time image simulation. This could give us access to invisible information contained within images, such as blood pressure which can be measured in vivo. Fluid animation could be used to track the flow of blood during an aneurysm, for example. At the moment, the treatment stops when the image shows that there is no longer any blood inside. A more intelligent system, featuring animation that is an accurate representation of the image - reproducing reality in space and time through data processing - would give us further arguments when it comes to planning and carrying out treatment."

What are the benefits of augmented reality?

E.K : "One way of improving image interpretation is to add three-dimensional information to the two-dimensional image the doctor sees. Augmented reality can be used to provide an intuitive summary of a range of complex data, helping doctors to make decisions without taking their place. At one point, the possibility of automating mammography diagnostics using computer-aided diagnosis was discussed. This, however, never happened.

The systems that are currently being developed indicate clearly where doctors should look. They are free to follow this advice or not, but it can save them time during examinations. In conclusion, my aim is for complex images, which are not always particularly informative or multimodal, to become synthesized and easy to interpret."

So, a 100% artificial image - does that mean it will be 100% effective?

E.K : "No. However, as computer scientists, after a number of years spent working with doctors, we are aware of their acknowledgement of the benefits of computer science and their willingness to incorporate this technology into their daily routines. For all of us, it is clear that each new intelligent innovation in health must still be subject to human medical approval."


© Re.Med. / Nov.2018

Laurence Verger

Research communication manager

Nancy CHRU (Regional University Hospital)