Image Analysis and Machine Learning Platform for Innovation in Diabetic Retinopathy Screening


ARSN is implementing a mass screening for diabetic retinopathy (DR), with the goal of making eye exam of about 75% of identified diabetics. The vision of the consortium SCREEN-DR is to create a distributed and automatic screening platform for DR, based on advanced PACS* management, Machine Learning and Image Analysis, enabling immediate response from health carers, allowing accurate follow-up strategies, and fostering technological innovation. As main objectives we have the automatic image quality assessment, the automatic detection and grading of diabetic retinopathy including the mild non-proliferative, moderate/severe non-proliferative and proliferative grades.


Main Task Goals:

Image Quality Assessment

To automatically reject low quality images that are not suitable for further analysis.

Detection of Normal Images

To automatically discriminate DR fundus images (regardless of DR type and stage) from ‘Normal’ ones.

DR Grading

To grade fundus images into three levels of severity: mild, moderate/severe and proliferative. These levels are characterized by the presence of several lesions, as microaneurisms, exudates, neovascularization, vessel tortuosity, hemorrhages, and venous beading.

Image Web Service

To provide a remote access to all the image analysis functionalities, namely for image quality evaluation, detection of image normality and DR grading.


Vessel Width Estimation demo at:

Convolutional Bag of Words demo at: