
Lirot.ai for retinal images analysis using artificial intelligence
Lirot.ai is a scholarly resource developed by the AIMLab. and committed to making AI-driven analysis of ophthalmology images widely available, simplifying the integration of AI in the study of ophthalmology and promoting equitable access to AI technology in the field for research and educational purposes.
A set of modules will be released over the next few months as part of our iOS Lirot.ai software and this page will be updated accordingly.
Recommended devices: Apple Silicon (M series) chip, iPad with Apple Pencil.
Formats supported: JPG, PNG and TIFF.
Contact: lirotai@technion.ac.il
Initial release date: 01-03-2025
05
AMD Identification Module
Deep learning generalization for Age-Related Macular Degeneration (AMD) identification from fundus images.
Deep learning generalization for AMD (Age-Related Macular Degeneration) identification from fundus images.Age-Related Macular Degeneration (AMD) is a leading cause of irreversible vision impairment worldwide, particularly among older adults. Early detection and monitoring are critical for managing AMD and preserving vision. The traditional diagnostic process for AMD involves a combination of fundus imaging and clinical assessment by a specialist, which can be resource-intensive and time-consuming. This module introduces AMDNet, a cutting-edge deep learning model developed using five independent datasets comprising more than 119,000 digital fundus images (DFI) with gold-standard annotations from diverse demographic backgrounds. AMDNet demonstrates exceptional out-of-distribution generalization with AUC ranging 0.87-0.95 across target domains. The performance of AMDNet surpasses traditional biomarkers, such as drusen size and retinal pigment abnormalities, by up to 20.5%.
Module features:
-
Probability of the eye being affected by AMD.
-
Explainability maps highlighting regions associated with AMD diagnosis.
-
Suitable for various DFI resolution and FOVs

We acknowledge the collaboration of Dr. Eran Berkowitz and Dr. Meishar Meisel.
Initial release date: 01-02-2025
04
Glaucoma Identification Module
Deep learning generalization for glaucoma identification from fundus images.
Glaucomatous optic neuropathy (GON) is a prevalent ocular disease that can lead to irreversible vision loss if not detected early and treated. The traditional diagnostic approach for GON involves a set of ophthalmic examinations, which are time consuming and require a visit to an ophthalmologist. This module implements GONet, a robust deep learning model developed using five independent datasets totaling more than 119,000 DFIs with gold-standard annotations and from diverse demographic backgrounds. GONet demonstrates high out-of-distribution generalization with AUC ranging 0.85-0.94 in target domains. The performance of GONet was superior to using cup-to-disc ratio by up to 21.6%.
Module features:
-
Probability of the eye to be glaucomatous.
-
Cup to disk ratio (CDR).
-
Cup and disk segmentation.
-
Explainability maps.
-
Suitable for various DFI resolution and FOVs

We acknowledge the collaboration of Prof. Eytan Blumenthal and Dr. Hadas Pizem
Initial release date: 26-11-2024
Last updated: 01-03-2025
03
Diabetic Retinopathy Staging Module
Deep learning generalization for diabetic retinopathy staging from fundus images.
Diabetic retinopathy (DR) is a direct microvascular complication of diabetes mellitus. Elevated glucose levels caused by diabetes mellitus lead to the production of cytokines and growth factors, resulting in damage to the capillaries of the eye's blood vessels. This damage increases vascular permeability and causes capillary occlusions. Early detection of DR is crucial, as any delay can lead to rapid vision degradation and eventual irreversible blindness. We developed DRStageNet, a deep learning algorithm for DR staging. DRStageNet was trained on over 90,000 fundus images from six independent datasets, encompassing a variety of demographics, ethnicities, geographic origins, comorbidities, image resolutions, and fields of view (FOVs).
Module features:
-
Severity classification of diabetic retinopathy (ICDR standard).
-
Explainability map.
-
Segmentation of microaneurysm, exudate and hemorrhage.
-
Suitable for various DFI resolution and FOVs.

We acknowledge the collaboration of Dr. Luis Filipe Nakayama.
Initial release date: 01-11-2024
Last updated: 01-03-2025
02
Digital Microvasculature Biomarkers Module
Computerized analysis of the eye microvasculature in mass datasets of digital fundus images.
The fundus image allows for the visualization of numerous vascular features, notably the arterioles (small arteries) and venules (small veins). Advanced image processing and machine learning techniques now enable the extraction of arterioles and venules from fundus images, a process known as A/V segmentation. By isolating these vessels, we can examine their morphology and distribution in greater detail, revealing subtle changes that might otherwise go unnoticed. From this A/V segmentation, we can compute vasculature biomarkers, which are quantifiable indicators of biological states or conditions. By analyzing these vasculature biomarkers, healthcare professionals can gain deeper insights into a patient’s ocular health and potentially detect early signs of disease.
Module features:
-
Digital vasculature biomarkers engineering (18 features).
-
Including a set of 12 biomarkers computed for arterioles (A) or venules (V).
-
Defining a region of interest.
-
Exporting digital vasculature biomarkers for subsequent analysis.
Remarks: All black borders are removed then the image is padded so to resize the image to a squared dimension. Finally, the image is resampled to 1444x1444 pixels. Biomarkers highlighted in blue can be computed for arterioles (A) or venules (V). All biomarkers are evaluated within a region of interest (ROI), which is by default the full 1444x1444 pixels image or can be selected to be a 500 pixels width circle around the optic disc.