Uni­ver­si­ty pro­fes­sor and emer­i­tus direc­tor of research (2017–2023), Marie-Chris­tine Ho Ba Tho spe­cial­izes in the devel­op­ment of per­son­al­ized mod­els of the mus­cu­loskele­tal sys­tem at UTC-BMBI. She leads the PREDIT4FACE project, one of six­teen win­ners of the PEPR San­té Numérique de France 2030 call for pro­pos­als, assigned a total bud­get of 24.5 Meuros.

Sched­uled to run for four years, PREDIT4FACE has a bud­get of 1.59 Meu­ros and aims to devel­op a mul­ti-scale pre­dic­tive dig­i­tal twin of the face. This ambi­tious project brings togeth­er, with­in a con­sor­tium of five part­ners, four of which are locat­ed in the Hauts-de-France region: Pro­fes­sor Stéphanie Dakpe, a max­illo­fa­cial sur­geon affil­i­at­ed with CHIMERE (UPJV-CHU Amiens, Inserm); Pro­fes­sor Tien Tuan Dao, a UTC alum­nus (BMBI), who has joined the UMR LaM­cube (Cen­trale Lille Insti­tut, CNRS); Pro­fes­sor Valérie Bur­din of Latim (IMT Atlan­tique, Inserm); Rober­val (UTC) with Pro­fes­sor Alain Rassineux, a spe­cial­ist in com­plex mesh­ing tech­niques; and Mohamed Rachik, an asso­ciate pro­fes­sor spe­cial­iz­ing in non­lin­ear behav­iour mod­el­ling; and final­ly the BMBI, which includes, in addi­tion to Marie-Chris­tine Ho Ba Tho, Olfa Tra­bel­si, an asso­ciate pro­fes­sor spe­cial­iz­ing in the char­ac­ter­i­za­tion and mul­ti­scale mod­el­ling of soft liv­ing tis­sues. This con­sor­tium brings togeth­er a vari­ety of com­ple­men­tary areas of expertise.

To devel­op a dig­i­tal twin of the face, mod­els must be cus­tomized both geo­met­ri­cal­ly and in terms of mechan­i­cal prop­er­ties. Var­i­ous tools for mod­el­ling, mesh­ing, med­ical imag­ing and arti­fi­cial intel­li­gence, as well as clin­i­cal exper­tise, are required. “Thus, using med­ical images, we devel­op a vir­tu­al 3D mod­el in which we recon­struct the face’s geom­e­try through seg­men­ta­tion. Next, we ‘mesh’ the face using finite ele­ments to run sim­u­la­tions. For this work, we rely in par­tic­u­lar on the exper­tise of our Rober­val col­leagues, a long-stand­ing col­lab­o­ra­tion,” explains Marie-Chris­tine Ho Ba Tho.

As pio­neers in patient-spe­cif­ic mod­el­ling, the researcher and her team aim, through PREDIT4FACE, to bet­ter under­stand, mod­el and reha­bil­i­tate facial expres­sion. “So far, we’ve tried to under­stand the role of the twen­ty or so mus­cles involved in facial expres­sions. We’ve sim­u­lat­ed the mus­cles used to smile, or those acti­vat­ed dur­ing cer­tain pro­nun­ci­a­tions such as the ‘‘O’ or ‘Pou.’ We’ve also devel­oped a tool to mon­i­tor func­tion­al reha­bil­i­ta­tion exercises.

With the PREDIT4FACE project, we want to take this a step fur­ther. Our main goal is to devel­op pre­dic­tive, mul­ti-scale dig­i­tal twins capa­ble of mod­el­ling, sim­u­lat­ing, and pre­dict­ing facial expres­sions.” These mod­els will incor­po­rate the mor­pho­log­i­cal and mechan­i­cal prop­er­ties of the mus­cles and soft tis­sues of the face at dif­fer­ent scales for each patient. “For each patient, we will extract their imag­ing data and com­bine it with our knowl­edge of bio­me­chan­ics to mod­el the behav­iour of each patient’s face specif­i­cal­ly,” she explains.

This project involves anoth­er major chal­lenge. “Rapid mod­el recon­struc­tion is a huge chal­lenge. To achieve this, part of the team is cur­rent­ly work­ing on automat­ing seg­men­ta­tion, while anoth­er part is focus­ing on auto­mat­ic mesh­ing. Fur­ther­more, the inter­ac­tion between the patient and their dig­i­tal twin must occur in real time. Specif­i­cal­ly, this will involve cap­tur­ing data and per­form­ing rapid recon­struc­tion so that, when the patient makes a facial expres­sion, the dig­i­tal twin inter­acts with the move­ment and sim­u­lates what the patient is doing. This pre­dic­tive dig­i­tal twin will also help us opti­mal­ly plan func­tion­al reha­bil­i­ta­tion. It will allow us to devel­op dif­fer­en­ti­at­ed sce­nar­ios and, for each patient, define the opti­mal move­ments for facial expres­sions,” she emphasizes.

This requires a large num­ber of inno­v­a­tive method­olog­i­cal devel­op­ments in mod­el­ling, finite ele­ment cou­pling, rein­force­ment learn­ing mod­els and mod­el reduc­tion. Today, the proof of con­cept has been estab­lished. “Also as part of the PREDIT4FACE project, DeepFaceX—a start­up cur­rent­ly being estab­lished by Pro­fes­sor Tien Tuan Dao—has devel­oped soft­ware called iTwin4Face, designed for the rapid recon­struc­tion and real-time analy­sis of facial mod­els,” she adds. Marie-Chris­tine Ho Ba Tho is also lead­ing — in col­lab­o­ra­tion with Pro­fes­sor Tien Tuan Dao — the SCAR_FACE project (Sim­u­la­tion and Per­son­al­ized Mechan­i­cal Char­ac­ter­i­za­tion of the FACE), a project select­ed fol­low­ing the call for pre-mat­u­ra­tion projects issued by the Sor­bonne Uni­ver­si­ty Alliance’s uni­ver­si­ty inno­va­tion clus­ter. “The goal is to devel­op a dig­i­tal tool capa­ble of visu­al­iz­ing the per­son­al­ized mechan­i­cal prop­er­ties of facial skin in vivo and in real time. This tool will build on doc­tor­al the­sis work con­duct­ed at BMBI on the mechan­i­cal prop­er­ties of skin (Tran Viet, 2005; Ang Xiao Fan, 2016) and on the dig­i­tal twin of the face (T. N. Nguyen, 2021). It was the work of the latter—who also won the Guy-Dénielou The­sis Prize in 2021—that led to the cre­ation of Deep­FaceX,” con­cludes Marie-Chris­tine Ho Ba Tho.

MSD