Grupo ULMA – Vision cameras, computers, industrial controller Xplanar
Grupo ULMA
Sector: Health
Business Case
Early detection of macular pathologies is crucial to prevent progression to advanced stages and reduce the risk of severe visual impairment, including blindness. Macular pathology affects the most important portion of the retina, where the fovea, responsible for central visual function, is located. Optical coherence tomography (OCT) has been established as the gold standard for early detection of these pathologies. This diagnostic method allows the visualization of retinal microstructures, which is essential for the control and follow-up of these conditions.
Objectives
Development of an OCT image classifier capable of identifying the presence of macular pathology. If abnormalities are detected, determine the presence of macular oedema. Classify the cause of macular oedema among Diabetic Macular Oedema (DME), age-related macular degeneration (AMD) and other causes. Anticipate the appearance of macular pathology symptoms in order to start treatment early. To reduce the costs associated with the treatment of patients in advanced stages of the disease through early detection. Evaluate the efficacy of treatment by monitoring progress.
Use case
OCT imaging is a cost-effective, safe and non-invasive procedure. AI4RET aims to improve the processes of detection and management of macular pathology. Its objective is to anticipate the onset of symptoms, allowing earlier treatment and thus improving treatment efficacy.
Infrastructure
On-Premise / Cloud / Edge
Technology
Automatic or Deep Learning
Data
OCT images with their respective diagnoses. Each image has been diagnosed by an ophthalmologist. The set has been diagnosed by 8 ophthalmologists.
Resources
Multidisciplinary team of: ophthalmologists and engineers with specific training in medical imaging, deep learning, MLOps, interoperability standards, medical device regulatory, front-end & back-end development. Infrastructure: Linux server with 512 GB storage, 64 GB RAM and dedicated 40 GB Nvidia A100 graphics card.
Difficulties and learning
Access to quality data; OCT images from DR screening programmes with their respective diagnoses, GDPR requirements and limited availability of specialised professionals for the diagnostic process of such images. Knowledge management of the multidisciplinary team.
KPIs (business impact and metrics of the model)
AI model for macular pathology detection in OCT. Sensitivity: 92% / Specificity: 78%
Funding
Shareholders’ equity
Collaborators, Partners
Germans Trias i Pujol University Hospital (HUGTIP). Germans Trias i Pujol Research Institute (IGTP).
