• There are no suggestions because the search field is empty.

Experimentica case study: accelerating preclinical analysis of ocular diseases

Scientists at the CRO Experimentica describe using AI to analyze Spectral Domain Optical Coherence Tomography scans to identify neovascular lesions.
Written by Aiforia
Maria Vähätupa PhD, Project Coordinator & Senior Scientist
Marc Cerrada-Gimenez, PhD, Director of in Vivo Pharmacology


What is your goal with using Aiforia?

Maria and Marc: We worked with Aiforia Create to design an algorithm that would allow for the analysis of Spectral Domain Optical Coherence Tomography (SD-OCT) scans to identify neovascular lesions on the mouse laser-induced choroidal neovascularization (CNV) model.

Tell us a bit about Experimentica:

Experimentica is a global CRO. We develop and offer an industry-leading portfolio of preclinical ocular models. Our mission statement is to bridge the gap between the development of novel ocular models and their use in developing clinical applications.

The company was founded in 2013 and is headquartered in Kuopio, Finland. We have since grown our operations globally to understand and support local and international pharma and biotech companies as well as academic teams.

Our senior staff originates from academia, so R&D excellence is our corporate culture. We have been involved in several international research programs to leverage our ability to innovate. This resulted in the development of unique in vivo, in vitro, and ex vivo models, as well as forefront capabilities in imaging and artificial intelligence.

You can find out more about Experimentica here.

Why did you decide to use Aiforia’s solutions?

Aiforia is one of the top companies implementing methodologies to develop specific AI-based algorithms applicable to the research that we are conducting.

What are you using Aiforia’s image analysis software for?

Interview with experimentica-1The mouse laser-induced CNV model is a preclinical model for the neovascular (wet) form of Age Macular Degeneration (AMD). AMD is the main cause of vision loss in people over 65 years. The CNV mouse model partially resembles the human version of the disease so it is a well established model and is widely popular.

The standard readouts of the CNV mouse model include the analysis of the leakage area from fluorescein angiography scans and analysis of the CNV lesions by histology at the endpoint. However, as choroid and retina are three dimensional tissues and the neovascular lesion grows as a 3D mesh it would be ideal to have a fast and accurate methodology for the in vivo quantification of the neovascular lesion volume.

What is the objective of this project? 

To identify an efficient and reproducible method to evaluate the volume of choroidal neovascular lesions from the CNV mouse model.

What is the AI model (or algorithm) you created with Aiforia used for?

It allows for the identification and analysis of neovascular lesions from SD-OCT scans of the mouse laser-induced CNV model.

How many images did you need for training your own AI model?

We ran several tests with various number of training images and modifying the algorithm parameters, such as region context size, complexity, iterations, etc. until the appropriate parameters were identified. The actual project included 101 annotated training images and 418 analyzed images.

Without Aiforia would you have been able to do this project?

We would have manually identified the lesion by using some kind of image analysis software. In comparison to Aiforia this method would need to be repeated for every single study.

However, now that we have a trained algorithm for the analysis of the images these can be uploaded and analyzed automatically. Moreover, as the parameters of the analysis are kept constant it will allow for a direct comparison of the results from different studies.

What was it like working with the Aiforia software?

The usage of the Aiforia software after the initial training becomes easy as the workflow follows the logic steps in designing and using the algorithm for the selected analysis.

What challenges did it let you overcome?

Aiforia software allowed for the easy design and training of an AI model tailor made to our analysis.

Did you find it easy to use the Aiforia software? 

After the initial training process where you learn where all the menus are located and their meaning, how to design and train the algorithm, then the software use became more natural.