A.I. systems are developed and trained on large, heterogeneous databases of images annotated by expert ophthalmologists. These graded datasets are used to train the model to identify patterns and anomalies indicative of retinal conditions.
After the training phase, to assess whether the system has reached performance comparable to that of human doctors, a testing phase is conducted using external datasets graded by ophthalmologists but not previously used in training
A.I. assessments are highly accurate and often comparable to the assessments made by ophthalmologists.
The accuracy of the screening model “with or without abnormalities” on testing phase showed 93% sensitivity (false negative rate: 7%) and 90.6% specificity (false positive rate: 9.4%).
The A.I.* solution integrated in VX 610 automatically detect specific abnormalities present in the retina from a list of 13 different retinal conditions (Diabetic retinopathy, Retinal vessel occlusion, Pathologic myopia, Exudative ARMD, Retinitis pigmentosa, Nonexudative ARMD, Epiretinal membrane, Optic disc atrophy, Central serous chorioretinopathy, Suspicious glaucoma, Macular hole, Retinal detachment, Retinal artery occlusion).
The A.I.* report not only displays the score of 4 major ones (Diabetic Retinopathy, Exudative ARMD, Nonexudative ARMD, Suspicious glaucoma), but also highlight the risk of any other abnormalities from the list.
Yes, the algorithm is continuously validated based feedback post-product launch.
Since the population distribution and the equipment in real screening scenarios may differ from those on which the algorithm has been developed, a continuous improvement of the algorithm based on new data and feedback is ensured. In case of any update, the corresponding EC marking change procedures are followed before bringing the algorithm to the market.
Yes, the A.I. system focuses on a 45° angle, covering the central retina without requiring pupil dilation.
An ophthalmologist can explore also the peripheral retina by dilating the pupil, which is beyond the current capability of the A.I. system.
If the A.I. system suspects a pathology, the A.I.* report shows a yellow indication and the risk score associated. A.I.* outcome must not be intended as a diagnosis, even in case of high-risk score.
When a suspect is detected, it is advisable to refer to an ophthalmologist requesting a human grading for further verification.
The A.I. assessment is immediate, providing results as soon as the image is processed.
Healthcare professionals need basic training on how to operate the A.I. software and interpret its results. This training ensures they can effectively integrate A.I. tools into their clinical practice knowing benefits and limitations.
If the image quality is insufficient, a new image can be acquired. If the quality remains inadequate, the patient should be referred to an ophthalmologist for an in-person examination.
Yes, all personal data is securely stored using advanced encryption protocols through our Visionix Nexus system, ensuring confidentiality and protection of patient information.
All personal data acquired within the European Union are stored and/or processed, after encryption, in Frankfurt.
With the growing demand for eye care, A.I. tools will aid eye care professionals in their daily practice by enhancing early detection, improving patient triage and prioritization, and accelerating routine screening and assessments. This will enable professionals to dedicate more time to complex cases, surgical procedures, personalized care, and patient consultations.
A.I. has the potential to augment the capabilities of ophthalmologists and other eye care professionals, enhancing their work rather than replacing them.
