Ophthalmology was one of the first fields to adopt and utilize Artificial Intelligence (AI) technology. Ophthalmic imaging lends itself to AI, and the industry has truly been ahead of its time in using algorithms to recognize disorders, namely retinal diseases, and glaucoma. However, using AI technology to detect and predict corneal disease has had a slower adaptation rate and there are fewer AI imaging libraries and models around anterior segment predictions. Developing a keratitis image ophthalmic image database is a crucial step towards improving eye care services and research.

KeraLink is partnering with multiple organizations around the world to collect keratitis images which we will develop a large image library from and can start developing algorithms. Our goal is to implement AI technology in low-middle-income-countries (LMIC’s) in order for a non-clinician to screen and identify patients to refer to a medical professional for a full diagnosis. When screening for keratitis, time is of the essence, and ensuring the patient receives the correct topical medical treatment for a bacterial or fungal infection is crucial. Having a patient being screened on the spot, is critical to help save the patient’s vision.

In LMIC’s blindness due to corneal disease presents at a much higher rate than in developed countries. This is due to several factors; minor eye injuries which may warrant going to a doctor to prevent the injury from developing into an ulcer or becoming infected, often are not treated before the injury worsens and it’s too late for the patient. This is often due to proximity to local medical care, cost of treatment or lack of knowledge about receiving care. When treatment and medical care is provided for eye injuries, timing is everything. Every hour is crucial to preventing vision loss. So, when a care provider does treat patients who present with an eye infection, they do not stop to first biopsy and determine whether the infection is bacterial, fungal or viral; rather patients are treated with both antibiotic and anti-fungal medications. However, this course of treatment is detrimental and causes more harm than good. Patients are developing antibiotic resistance, and treatment with incorrect medications, in some cases, may cause the infection to worsen.

Identifying and determining the correct infection as soon as possible so the proper medications can be administered is a key step towards helping save vision. By developing this image library and creating the algorithm to distinguish between bacterial and fungal keratitis, we will be providing a tool to screen and diagnose patients earlier, which in turn may save their vision.

This database library will aid in the diagnosis of infective keratitis and facilitate early detection, which is critical for effective treatment.

Methodology:

The collection of ophthalmic images will be done in collaboration with ophthalmologists, KeraLink staff, and other eye care professionals. Several organizations across India have initially agreed to share their images with KeraLink International.

We will collect 5000 images of each: bacterial, fungal, and viral keratitis.  These will be downloaded along with patient demographic information, the data will be anonymized and then exported where it can be crafted to teach the AI.

The collection of a database of keratitis images is a critical step towards improving eye care services in LMICs and reducing vision loss from corneal disease. Moreover, the database will serve as a valuable resource for eye care professionals, researchers, and educators.  To date, there are few open-source databases which focus solely on anterior segment.  This will contribute to advancing knowledge of this subset of ophthalmology.  The collection of these images will require collaboration between organizations, eye care professionals and funding agencies to ensure the availability of high-quality images as well as appropriate technology to decipher and create the artificial intelligence to create the algorithms.

Top Image Credit: Jkokavec, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons