Article published Automated diagnosis
Learning: a strategic tool for decision making
The LIST is looking for industrial partners to develop intelligent systems for automatic diagnostics in the areas of security, predictive maintenance, personalised medicine and ambient intelligence.
What do these various devices have in common? An alarm system which rapidly locates earthquakes or anticipates industrial breakdowns, an algorithm which predicts sudden variations in the price of electricity on international markets, an expert system that assists in diagnosis and therapeutic choices for complex illnesses such as cancer, an automated assistance system for dependant people, etc. They are all able to take strategic decisions on complex subjects, by integrating multiple pieces of information.
“Beyond our academic research work, we are very familiar with real world problems and we are working on solutions for them in close collaboration with our industrial partners. We are able to understand their problems and offer them suitable solutions”, explains Jean-Denis Muller, manager of the Multisensors and learning intelligence team at CEA LIST.
Whatever the application, the approach is always the same. Information is collected by sensors, whether in a factory, vehicle, or electrical network; from seismic sensors or from measurements on the human body (biological, genetic). These data are then programmed, or in other words processed to increase efficiency, and included in algorithms which are able to supply the best diagnosis. “It is here that learning plays a role", explains Jean-Denis Muller. "Our algorithms learn how to make decisions about complex situations, rather like a medical student who perfects his expertise over the course of the cases that his professors present to him. Our systems undergo a training phase before they are declared fit for service.” The team has demonstrated its ability, for example, to handle highly complex problems from systems requiring a high level of reliability, developed for the security and intelligence industries. Amongst these is the field of predictive maintenance, which enables early detection of situations likely to lead to a breakdown and which are developing bit by bit. “We are also looking to apply our expertise in terms of personalised medicine, where there are a very large number of possible applications”, added the researcher. In collaboration with a hospital team, the laboratory has, for example, developed an expert system to help in diagnosis and therapeutic choices for urological cancers. This could be extended to other families of complex illnesses. The team is also working on prediction of individual risk, specified on the basis of genetic profiles, as well as on very rapid calculation systems to control radiotherapies in real time. Other emerging fields include ambient intelligence, an example of which is provision of assistance to dependent persons using a network of sensors which are not very invasive on their environment. This makes it possible to detect abnormal behaviour such as movement difficulties, a fall etc.
