Pro­fes­sor Marie-Chris­tine Ho Ba Tho coor­di­nates, for UTC the FIGURES team led by Pro­fes­sor Devauchelle of the Amiens Uni­ver­si­ty Hos­pi­tal, which per­formed the world’s first face trans­plant in 2005.

Pio­neers of patient-spe­cif­ic mod­els, the team of Marie-Chris­tine Ho Ba Tho aims to cre­ate the con­di­tions for opti­mal facial reha­bil­i­ta­tion fol­low­ing a trans­plant or facial paral­y­sis, work­ing with Stéphanie Dakpé, max­illo­fa­cial sur­geon. “The clin­i­cal team has observed a lack of under­stand­ing of facial expres­sions. It is dif­fi­cult to grasp the role of the facial mus­cles, of which there are almost twen­ty, which are acti­vat­ed accord­ing to face move­ments”, she explains.

How can we help them solve this prob­lem? “With our exper­tise, we mod­elled the face using the finite ele­ment method, using the sub­jec­t’s geo­met­ric and phys­i­cal data to sim­u­late facial expres­sions. We still had to define the facial move­ments in the case of facial reconstruction/rehabilitation,” explains Marie-Chris­tine Ho Ba Tho.

To out­line a solu­tion, she made the deci­sion – a world first – to com­bine finite ele­ment mod­el­ling of the face with deep learn­ing. “The for­mer enabled us to vir­tu­al­ly recon­struct the face and to sim­u­late the func­tion­ing of the mus­cles and their role in the case a smile, for exam­ple, or even pro­nun­ci­a­tions such as the ‘O’ or the ‘Pou’ used for facial reha­bil­i­ta­tion. The sec­ond is that of deep, gen­er­a­tive learn­ing. This involves teach­ing an agent to make sequen­tial deci­sions while inter­act­ing with an envi­ron­ment. The agent will be the patien­t’s face and the envi­ron­ment the finite ele­ment mod­el. Fol­low­ing a face trans­plant or facial paral­y­sis, we ar faced with the ques­tion of sim­u­lat­ing the mus­cles to be acti­vat­ed to achieve the desired result, such as restor­ing a smile or a sym­met­ri­cal face in the case of facial paral­y­sis,” she says.

Final­ly, the third stage con­sists of devel­op­ing com­ple­men­tary tools to refine the mod­el by super­im­pos­ing the tex­ture of the patien­t’s face using cam­eras. A start-up is being cre­at­ed for this pur­pose with Pro­fes­sor Dao. “We have thus devel­oped a tool that uses a patient-spe­cif­ic mod­el cou­pled with AI to sug­gest mus­cle actions lead­ing to the desired result. We test­ed it on the case of an asym­met­ri­cal face and val­i­dat­ed it by com­par­ing its mus­cle rec­om­men­da­tions for per­form­ing a par­tic­u­lar move­ment with those of the finite ele­ment mod­el. With this dig­i­tal tool, we have in short achieved a “proof of con­cept”,” con­cludes Marie-Chris­tine Ho Ba Tho.

MSD