Projects

Top Sector Life Sciences & Health (LSH) entails a broad scope of disciplines, from pharmaceuticals to medical technology and from healthcare infrastructure to vaccination. To realise its mission – vital citizens in a healthy economy - the Top Sector builds on the strengths of the Dutch LSH sector to address the biggest societal challenges in prevention, cure and care. By funding multidisciplinary public-private partnerships (PPPs) the Top Sector aims to facilitate innovation. Here we give an overview of  a number of funded R&D projects by Top Sector LSH. The page is updated continuously.

DEEPTISSUE: deep learning for cardiac tissue classification

Deep Learning for fully automated cardiac tissue classification from dynamic cardiac MR and CT imaging

Recent advances in MR and CT enable a precise assessment of the severity of heart diseases by showing how well the heart muscle is perfused. This can help in the early diagnosis of these diseases, and in the better planning of therapy. In this project, LUMC and Medis will develop automated analysis software that helps the radiologist interpret the patient’s condition from these images.

Over the past 4 years, a novel technology known as “Deep Learning” has revolutionised the field of automated image analysis and object recognition. However, the amount and complexity of such dynamic MR and CT data requires “artificially intelligent” techniques that help the radiologist and cardiologist to interpret this data. Therefore the aim is to develop framework based on Deep Learning for tissue classification from these complex MR and CT scans. 

Ultimately, the proposed software will help the medical professional with early diagnosis (before the damage is done), with the selection of the proper treatment (intervention). This will contribute to more personalised therapy, with benefits for the patient and the medical professional.

Results

Recent advances in MR and CT enable a precise assessment of the severity of heart diseases by showing how well the heart muscle is perfused. This can help in the early diagnosis of these diseases, and in the better planning of therapy. However, the amount and complexity of such dynamic MR and CT data requires “artificially intelligent” techniques that help the radiologist and cardiologist to interpret this data. Over the past 4 years, a novel technology known as deep learning has revolutionized the field of automated image analysis and object recognition. 

The aim of this proposal was to further develop this technology into a framework for “artificially intelligent” tissue classification from these complex MR and CT scans. Together with Medis, we aimed to develop automated analysis software that helps the radiologist interpret the patient’s condition (early diagnosis, before the damage is done). Alternatively, it can help the cardiologist to select the proper treatment (intervention). This will contribute to more personalized therapy, with benefits for the patient and the medical professional.  

The project resulted in a number of novel machine learning technologies to analyse time-varying cardiac MR and CT data. We developed a manifold-learning-based technique HSNE to cluster the image data into patches of similar temporal characteristics, and we developed a neural network-based approach to generalize this technique to larger data volumes. We also demonstrated that this approach gives good results in cardiac MR T1 mapping. As such, the original project goals were achieved; next steps outside the scope of this project is testing of the developed methodologies in multiple clinical centers, to demonstrate generalisability of the developed technologies.

See the webiste of Ikeb for more information about the project.