Home / Chroniques / New algorithms predict eyesight loss caused by ageing 
π Health and biotech π Digital

New algorithms predict eyesight loss caused by ageing 

Maëlle Vilbert
PhD student in physics at LOB* (IP Paris) and Centre Hospitalier National d'Opthalmologie des Quinze-Vingts
Key takeaways
  • Opacification of the eye is a phenomenon that affects about 10 million people worldwide.
  • Today, clinical evaluation tools do not allow for early diagnosis or quantitative monitoring of corneal pathologies: the doctor must interpret the result.
  • Maëlle Vilbert is developing an efficient image analysis method to avoid potential interpretation bias.
  • Artificial Intelligence (AI) models, if properly trained, can detect problems that humans would not see with the naked eye.
  • Using AI would allow doctors to spot potentially pathological situations very quickly and ensure better patient care.

“The World Health Organ­i­sa­tion [WHO] esti­mates that 80% of blind­ness is pre­ventable,” explains Maëlle Vil­bert, a doc­tor­al stu­dent at École Poly­tech­nique (IP Paris). More than 10 mil­lion peo­ple in the world are affect­ed by a visu­al hand­i­cap due to the loss of trans­paren­cy in the cornea. Although, this issue remains less com­mon than cataracts (opaci­fi­ca­tion of the crys­talline lens), and glau­co­ma (linked to exces­sive intraoc­u­lar pres­sure), loss of corneal trans­paren­cy remains one of the main sources of dete­ri­o­ra­tion, or even total loss, of vision.

The eye is one of the most com­plex organs in the human body. For it to func­tion prop­er­ly, and there­fore for clear vision, it must be com­posed of sev­er­al healthy com­po­nents. How­ev­er, as we age, these ele­ments dete­ri­o­rate in many peo­ple. Thus, even if the caus­es of blind­ness are diverse and var­ied, one of the main caus­es is the opaci­fi­ca­tion of the cornea. This nat­u­ral­ly trans­par­ent lens, which cov­ers the iris at the front of the eye, allows light to pass through. Its cloud­ing direct­ly affects a person’s vision – a phe­nom­e­non that affects about 10 mil­lion peo­ple worldwide. 

Opaci­fi­ca­tion of the eye is a phe­nom­e­non that affects about 10 mil­lion peo­ple worldwide.

Today, corneal trans­plan­ta­tion is the most com­mon type of trans­plant in the world. And whilst, it remains unavoid­able to treat advanced stages of corneal opaci­fi­ca­tion, it is bet­ter to pre­vent it entire­ly. Indeed, aside from the risks asso­ci­at­ed with the oper­a­tion, there is a seri­ous short­age of corneal grafts world­wide, with an aver­age of 1 donat­ed cornea for every 70 needed. 

Accord­ing to Maëlle Vil­bert, the clin­i­cal tools for assess­ing corneal trans­paren­cy remain qual­i­ta­tive and/or oper­a­tor-depen­dent, which does not allow for ear­ly diag­no­sis or quan­ti­ta­tive mon­i­tor­ing of corneal patholo­gies. “Prac­ti­tion­ers analyse opti­cal coher­ence tomog­ra­phy (OCT) images with the naked eye,” explains the researcher. “There is no stan­dard­ised method for extract­ing prop­er­ties direct­ly relat­ed to the tis­sue. This leaves room for the doc­tor’s sub­jec­tiv­i­ty. If the prob­lem is not obvi­ous, he or she may not be able to see it.”

“Even so, this imag­ing method records the image of each cornea that has been exam­ined,” explains Maëlle Vil­bert, “which pro­vides us with an enor­mous data­base”. Vil­bert is work­ing on this data for her the­sis to devel­op a method of image analy­sis that will enable a phys­i­cal mea­sure­ment of corneal trans­paren­cy, so as to avoid poten­tial bias­es in the inter­pre­ta­tion of images. 

“By understanding its transparency, we understand its opacification”

Trans­par­ent tis­sue is unusu­al in nature, but this char­ac­ter­is­tic of the cornea can be explained. The stro­ma is a con­nec­tive tis­sue that makes up 90% of the thick­ness of the cornea. It is com­posed of nanoscale col­la­gen fib­rils whose diam­e­ter and spac­ing with­in strat­i­fied lamel­lae reflect a localised order­ly organ­i­sa­tion. This local order gives rise to destruc­tive inter­fer­ence in the tis­sue in all spa­tial direc­tions except direct trans­mis­sion, hence the remark­able trans­paren­cy of the cornea. Only the light sig­nal trans­mit­ted direct­ly through the cornea and lens allows images to be formed on the retina. 

Organ­i­sa­tion of col­la­gen fib­rils in a nor­mal cornea, in an oede­ma­tous cornea and in the scle­ra. Repro­duced from [Pla­mann et al., 2010]. Image tak­en from Maëlle Vil­bert’s the­sis, “In vivo opti­cal diag­no­sis of corneal trans­paren­cy by opti­cal coher­ence tomog­ra­phy (OCT)”.

“A light wave can be trans­mit­ted, absorbed, or scat­tered by a medi­um,” says Maëlle Vil­bert, “and the cornea does not absorb any­thing, so it either trans­mits or scat­ters the light. As soon as the scat­ter­ing phe­nom­e­na become too great, the cornea becomes opaque, and its trans­paren­cy is lost.” Hence, it is this order­ly organ­i­sa­tion of the col­la­gen fib­rils mak­ing up the stro­ma that makes corneal trans­paren­cy pos­si­ble. If its com­po­si­tion becomes dis­or­dered, as is the case in the scle­ra – the white of the eye – where the diam­e­ter of the col­la­gen fib­rils is not con­stant, light is no longer trans­mit­ted direct­ly into the eye. 

“Corneal oede­ma is one of the caus­es of opaci­fi­ca­tion,” she adds, “because it caus­es the appear­ance of micro­met­ric aque­ous inter­stices between the col­la­gen lamel­lae of the stro­ma, often called ‘lakes’, which scat­ter the light.”

Coupling physics and AI for a more reliable and accurate analysis 

Maëlle Vilbert’s project is based on the hypoth­e­sis of a homo­ge­neous corneal stro­ma in order to char­ac­terise its trans­paren­cy using phys­i­cal para­me­ters. “A het­ero­ge­neous envi­ron­ment would cause local vari­a­tions in the atten­u­a­tion of the OCT sig­nal,” Vil­bert explains. “By sta­tis­ti­cal­ly val­i­dat­ing the con­sis­ten­cy of the corneal stro­ma, we can quan­ti­fy its trans­paren­cy using a sin­gle per­cent­age of trans­mit­ted coher­ent light. The aim is to stan­dard­ise meth­ods for analysing OCT images. This allows us to dis­tin­guish between a nor­mal cornea and a patho­log­i­cal cornea with low scat­ter­ing, which is dif­fi­cult to detect with cur­rent clin­i­cal tools.” 

How­ev­er, corneas with localised opac­i­ties can­not pro­vide a sin­gle trans­paren­cy para­me­ter. “We have also adopt­ed an approach of auto­mat­ic clas­si­fi­ca­tion of clin­i­cal images to detect cer­tain corneal inflam­ma­tions, such as in Fuchs’ dys­tro­phy or corneal haze after refrac­tive surgery. AI mod­els, if prop­er­ly trained, can detect prob­lems that humans would not see with the naked eye.”

“Of the var­i­ous para­me­ters used by the team to train the AI mod­el, one para­me­ter (“sig­ma”) alone has a clas­si­fi­ca­tion accu­ra­cy of 93%: it reflects the depth of the fire zone. The oth­er 8 para­me­ters increase the clas­si­fi­ca­tion accu­ra­cy to 97%. These para­me­ters can still be inter­pret­ed by prac­ti­tion­ers,” she insists, “which is essen­tial for the pos­i­tive recep­tion of this type of dig­i­tal diag­nos­tic tool.” Doc­tors could thus use this AI to detect cer­tain symp­toms at an ear­ly stage, espe­cial­ly when they are not vis­i­ble to the naked eye and ensure fol­low-up over time for bet­ter patient care.

The meth­ods devel­oped by this team are com­ple­men­tary tools to tra­di­tion­al slit lamp and OCT diag­no­sis. Being able to assess the trans­paren­cy of the cornea with such accu­ra­cy in the giv­en per­cent­age allows poten­tial­ly patho­log­i­cal con­di­tions to be iden­ti­fied very quick­ly. It also makes it pos­si­ble to estab­lish an effec­tive fol­low-up over time, as it is quan­ti­ta­tive. Thus, it would be pos­si­ble to inter­vene ear­ly and avoid the need for inva­sive treat­ments such as a corneal transplant. 

AI mod­els, if prop­er­ly trained, can detect prob­lems that humans would not see with the naked eye.

“There is a real chal­lenge in this analy­sis tech­nique,” explains Maëlle Vil­bert, “giv­en the age­ing pop­u­la­tion and the fact that 80% of cas­es of blind­ness are pre­ventable. More accu­rate diag­no­sis, both at the time and in the long term, means more effec­tive pre­ven­tion and bet­ter patient care.” The tool designed by the researcher and her team is easy to use due to its automa­tion. Thus, after a short train­ing course, the diag­no­sis of corneal trans­paren­cy is afford­able for peo­ple who are not experts in the field.

These meth­ods cou­pling AI with the physics of light prop­a­ga­tion in human tis­sues have a promis­ing poten­tial for the design of clin­i­cal diag­nos­tic tools. They could, for exam­ple, be trans­ferred to the diag­no­sis and mon­i­tor­ing of cataracts, a con­di­tion that accounts for more than half of all visu­al impair­ments worldwide.

Pablo Andres

Fur­ther reading 

For more details on the research, here is the work of Maëlle Vil­bert: https://www​.the​ses​.fr/s242473

Our world explained with science. Every week, in your inbox.

Get the newsletter