Karim El Kirat, a uni­ver­si­ty pro­fes­sor, is co-head, with Sofi­ane Boudaoud, of the C2MUST team at UTC-CNRS-BMBI. He devel­ops mul­ti-scale mod­els for fall pre­ven­tion, using AI in particular.

It was with Tuan Dao, then a mem­ber of UTC-BMBI and an AI spe­cial­ist, that Karim El Kirat began work­ing on mul­ti-scale mod­els at the mol­e­c­u­lar lev­el. Since then, Tuan Dao has joined the Ecole Cen­trale de Lille as a pro­fes­sor, but their col­lab­o­ra­tion has con­tin­ued on mul­ti-scale mod­el­ling of the whole body. “The bio­me­chan­ics of the human body inte­grates dif­fer­ent aspects. We are inter­est­ed in the mus­cu­loskele­tal sys­tem, name­ly the mus­cles, bones and ten­dons. The objec­tive is to define the role of the dif­fer­ent ele­ments dur­ing a move­ment, for exam­ple. The bone is the foun­da­tion of the sys­tem, the ten­dons are springs of a sort, while the mus­cles are actu­a­tors. The ques­tion is: who con­trols all this and how does it work? When we do mul­ti-scale mod­el­ling, we simul­ta­ne­ous­ly analyse how the mus­cle will con­tract at dif­fer­ent scales to pro­duce move­ment on the human scale, and how nerve impuls­es par­tic­i­pate in all this,” explains Karim El Kirat.

The two research sci­en­tists were par­tic­u­lar­ly inter­est­ed in the prob­lem of falling. “When we’re small, we learn to walk by falling over, get­ting up and con­tin­u­ing to walk, or so we think, for life. How­ev­er, sar­cope­nia, the infil­tra­tion of fat into the mus­cle that low­ers mus­cle strength, can begin as ear­ly as the age of 40, lead­ing to an increas­ing risk of falling with age, espe­cial­ly for seden­tary sub­jects. Thus, in the mod­el­ling, it was decid­ed to use a bio­mimet­ic AI that mim­ics the process of learn­ing to walk. A mod­el based on a spe­cif­ic AI was devel­oped: gen­er­a­tive deep learn­ing. A bio­me­chan­i­cal mod­el of the whole body is cre­at­ed, inte­grat­ing the mechan­i­cal prop­er­ties of bone, mus­cle and ten­don. The mod­el is then asked to take a few steps and to fall. The mod­el is reward­ed if it achieves the objec­tives (falling, or not) and it is thus pos­si­ble to analyse which parts of the body, which stages of walk­ing are deci­sive if we want to avoid falling. The mod­el is trained using the data avail­able in our data­base but also data from the lit­er­a­ture in order to ulti­mate­ly pro­pose pre­ven­tive mus­cu­loskele­tal strength­en­ing rou­tines,” he concludes.

MSD