Deep Learning for early detection of diabetic retinopathy

Diabetic retinopathy

Diabetic retinopathy (DR) is the prime cause of preventable blindness in adults of working age. It affects 4.1% of Europeans, or 1 out of every 3 persons with diabetes mellitus (DM), habitually type-2 diabetes, which is the commonest kind. About 2% of DM patients will turn blind after 15 years and 80% will develop some type of DR. These figures are important because they are all DM-related, and DM is one of the world’s most widespread illnesses.

DR-caused vision loss is sometimes irreversible but early detection by means of periodic checks can cut the blindness risk by 95%. The current detection system is inefficient, mainly for the following reasons:

1) Many patients skip some or even all of the planned visits due to ignorance about the importance of regular screening.

2) The checks are always carried out by ophthalmologists; this is costly and time-consuming, lengthening waiting lists.

3) It calls for dilated eye exams, which are an uncomfortable procedure.

4) A high percentage of cases (20-30%) fly under the radar.

Indeed, current solutions are based on DR detection by ophthalmologists. Most regions lack widespread screening programs because they are very costly, time-consuming and inadequate; the dilated-eye exams involve the use of expensive mydriatic cameras (retinographs). Primary healthcare facilities, the main point of contact for DM patients, lack the equipment, personnel skills and, above all, the time for exams of this type. The upshot is a large amount of late-detected or even non-diagnosed DR sufferers. This means that an illness that could be treated simply ends up costing a lot per patient, since, once the illness is advanced, the treatment usually calls for surgery.

One alternative is to switch the diagnosis to non-specialized centers or actors, i.e. to free ophthalmologists from this task. Early DR-diagnosis techniques based on deep learning achieve high precision, using fundus cameras that do not need eye dilation (called non-mydriatic). Cameras of this type have a narrower field of view, about 40º as opposed to practically the whole of the retina in mydriatic retinography. Even so, deep-learning algorithms achieve a precision of between 80% and 92%. Equipping primary healthcare with this capacity would substantially reduce referrals to ophthalmology but, above all, it would increase the number of sufferers diagnosed.

GMV’s healthcare division boasts a wealth of experience in healthcare innovation in various sectors, with a wide range of inhouse products taking in telemedicine, tele-monitoring or illness-predicting AI; the company also takes part in many innovation projects at both national and international level. Drop into our website to find out about our healthcare projects.

Author: Javier Téllez Chacón

Las opiniones vertidas por el autor son enteramente suyas y no siempre representan la opinión de GMV
The author’s views are entirely his own and may not reflect the views of GMV
Share

    Leave a Reply

    Your email address will not be published. Required fields are marked *

      I accept the privacy policy Acepto la Política de Privacidad

    Basic Data-Protection information:

    Data-protection supervisor: GMV Innovating Solutions SL
    Purpose: Answer questions, respond to user complaints and recommendations, receive job applicants resumes and career information.
    Legitimation: Consent of data subject
    Addresses: Grupo GMV companies
    Rights: Access, rectify and cancel data plus other rights, as explained in additional information
    Additional information: You can check out the additional and detailed data-protection information on our website:Privacy Policy

    You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

    WP-SpamFree by Pole Position Marketing

    765 Total Views