Translating AI research to real use cases

Hello and welcome! Some of you may know us well, and some of you may have only heard of us and our work. We want to welcome you to our first newsletter which will inform you and hopefully inspire you with all things related to AI in medical image analysis.

AI for Oncological Imaging Analysis

Benjamín Gutierrez Becker , R&D Imaging Data Analysis Expert at 罗氏公司 , gave a speech titled ‘AI for Oncological Imaging Analysis’ conference on AI pharmaceutical projects in which we participated at the #BioTechX . You could learn about automated tumor burden measurement from MRI and automated liver cirrhosis detection in CT scans. For more details please check our website (cirrhosis project , pre-operative glioma project , post-operative glioma project ).

On the very same conference Bartosz Machura , our medical imaging expert, presented a poster entitled “An end-to-end pipeline for automating multi-class brain tumor segmentation and volumetric measurements” and explained how AI helps in automatic progression analysis and disease tracking. If you missed it, don’t worry – we've?got you covered and published the poster here .?

You have to compete against the best

The past weeks have been very busy for our medical imaging experts. ?We entered two competitions with our brain tumor algorithm: the Federated Tumor Segmentation Challenge 2022 (FeTS) and the Brain Tumor Segmentation Challenge (BraTS).?We are pleased to announce that Graylight Imaging team (Jakub Nalepa ,?Krzysztof Kotowski ,?Wojciech Malara ,?Bartosz Machura ,?Szymon Adamski ) has achieved unprecedented success?by placing right behind the winners in second place in FeTS and second place as well in the pediatric population of diffuse intrinsic pontine glioma patients dataset and third in the underrepresented Sub-Saharan Africa adult patient populations of brain diffuse glioma dataset in BraTS. We are honored to have received such recognition for our work, which was up against models developed by world-class experts in AI-enabled medical image analysis and tumor segmentation. Congrats to all participants! Read more

Coronary arteries and Deep Learning puzzles

When we say that the past weeks have been busy, it is no joke. At the EuSoMII – European Society of Medical Imaging Informatics Annual Meeting in Valencia our deep learning expert Jaroslaw Goslinski, PhD had an opportunity to share our experiences on automated coronary arteries segmentation. The topic of the presentation was as follows: “The size matters: Compensating the impact of small patches on 3D U-Nets for precise segmentation of coronary vessels” and during the speech we answered the basic question: how to choose the proper size of the patch when building the models for the 3d segmentation.

Coronary arteries segmentation and analysis issues is a topic we have taken up on yet another occasion in recent weeks. At?#MICCAI2022 ?our team represented by?Filip Malawski ?had an oral presentation on how Deep Learning meets Computational Fluid Dynamics to assess coronary artery disease (CAD) in CCTA. Our experiments, performed over clinically acquired scans, revealed that the segmentation approaches suggested by us not only outperform state-of-the-art nnU-Nets, but also lead to the blood-flow parameters which are in strong agreement with those elaborated for the ground-truth delineations. Over 0.8 - it is the average S?rensen-Dice coefficient of the obtained by us models. The accuracy of coronary artery tree reconstruction exceeds 90%.

This rigorous study was published: Malawski, F.?et al.?(2022). Deep Learning Meets Computational Fluid Dynamics to?Assess CAD in?CCTA. In: Wu, S., Shabestari, B., Xing, L. (eds) Applications of Medical Artificial Intelligence. AMAI 2022. Lecture Notes in Computer Science, vol 13540. Springer, Cham. Deep Learning Meets Computational Fluid Dynamics to?Assess CAD in CCTA

This interest in the cardiac area is related to our internal R&D project on coronary artery segmentation, the effects of which we plan to present in the coming months – stay tuned!

Sounds interesting? Would you like to chat with our experts to learn more about the medical imaging algorithms we develop? Get in touch and pm Przemys?aw Urbański or drop him a line at [email protected].