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
Release date: 01-05-2024
01
LUNet Module
Deep learning for the segmentation of arterioles and venules in high resolution fundus images.
The retina is the only part of the human body in which blood vessels can be accessed non-invasively using imaging techniques such as digital fundus images (DFI). The spatial distribution of the retinal microvasculature may change with cardiovascular diseases and thus the eyes may be regarded as a window to our hearts. Computerized segmentation of the retinal arterioles and venules (A/V) is essential for automated microvasculature analysis. We developed LUNet, a novel deep learning architecture for high resolution A/V segmentation The LUNet module also provides a user-friendly interface for editing the automated segmentations.
Module features:
-
Segmentation of arterioles (A) and venules (V).
-
User interface for manual correction.
-
Optic disk and cup segmentation.
This module is released in collaboration with Prof. Ingeborg Stalmans and Dr. Jan Van Eijgen from KU Leuven, Belgium. We also acknowledge the contributions of Dr. Moti Freiman and Motti Koren at the Technion.
Release date: 01-11-2024
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.
This module is released in collaboration with Prof. Ingeborg Stalmans and Dr. Jan Van Eijgen from KU Leuven, Belgium.
Release date: 01-12-2024
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, Dr. Leo Anthony Celi.
Associated Publications
-
LUNet: deep learning for the segmentation of arterioles and venules in high resolution fundus images. Fhima J, Van Eijgen J, Billen Moulin-Romsée MI, Brackenier H, Kulenovic H, Debeuf V et al. LUNet: deep learning for the segmentation of arterioles and venules in high resolution fundus images. Physiological Measurement. 2024 May 3;45(5). [DOI] [Link to publication in Scopus]
-
Leuven-Haifa High-Resolution Fundus Image Dataset for Retinal Blood Vessel Segmentation and Glaucoma Diagnosis. Van Eijgen J, Fhima J, Billen Moulin-Romsée MI, Behar JA, Christinaki E, Stalmans I. Leuven-Haifa High-Resolution Fundus Image Dataset for Retinal Blood Vessel Segmentation and Glaucoma Diagnosis. Scientific data. 2024 Dec;11(1):257. [DOI] [Link to publication in Scopus]
-
PVBM: A Python Vasculature Biomarker Toolbox Based on Retinal Blood Vessel Segmentation. Fhima J, Eijgen JV, Stalmans I, Men Y, Freiman M, Behar JA. PVBM: A Python Vasculature Biomarker Toolbox Based on Retinal Blood Vessel Segmentation. In Karlinsky L, Michaeli T, Nishino K, editors, Computer Vision – ECCV 2022 Workshops, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 296-312. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). [DOI] [Link to publication in Scopus]
-
Computerized analysis of the eye vasculature in a mass dataset of digital fundus images: the example of age, sex and primary open-angle glaucoma. Fhima J, Eijgen JV, Reiner-Benaim A, Beeckmans L, Abramovich O, Stalmans I, Behar JA. medRxiv (2024): 2024-07. [DOI]
-
DRStageNet: Deep Learning for Diabetic Retinopathy Staging from Fundus Images. Yevgeniy M, Fhima F, Celi LA, Ribeiro LZ, Nakayama LF, Behar JA. arXiv preprint arXiv:2312.14891 (2023). [DOI]