DeepNeo achieves the precision of medical professionals while significantly reducing evaluation time. Thanks to extensive validation in human and animal models, DeepNeo has the potential to standardise follow-up care after stent operations - and thus sustainably improve treatment outcomes for cardiovascular diseases.
Challenges in monitoring the stents
Every year, more than three million people worldwide are treated with stents to widen blood vessels narrowed by heart disease. However, monitoring the healing process after implantation remains a challenge. If the tissue growing over the stent develops irregularly - either too thick or with debris - this can lead to complications such as re-narrowing or blockage of the blood vessel. Currently, analysing these healing patterns in intravascular optical coherence tomography (OCT) images is time-consuming and difficult to implement in routine clinical practice.
Automated assessment of the healing process
A research team from Helmholtz Munich and the TUM Klinikum has developed DeepNeo, an AI algorithm that can automatically assess the healing of stents in OCT images. DeepNeo recognises different healing patterns with an accuracy equivalent to that of clinical experts - but in a fraction of the time. In addition, the AI tool delivers precise measurement data, such as tissue thickness and stent coverage, providing valuable insights for patient management.
‘With DeepNeo, we achieve an automated, standardised and extremely precise analysis of stent and vessel healing - something that was previously only possible through time-consuming manual evaluation,’ says Valentin Koch, first author of the study in which the algorithm was presented. ‘DeepNeo is as good as a doctor - only much faster.’
Validated with strong performance
To train DeepNeo, the research team used 1,148 OCT images from 92 patient scans, which were manually annotated to classify different forms of tissue growth. The AI algorithm was then tested in an animal model - with convincing results: DeepNeo correctly identified pathological tissue in 87 per cent of cases compared to detailed laboratory analysis, the current gold standard. DeepNeo also demonstrated a high level of precision when analysing human scans and was in close agreement with the assessments of medical professionals.
‘DeepNeo shows how machine learning can help doctors make faster and more informed treatment decisions. The next step is now to integrate AI algorithms like DeepNeo into clinical practice in a targeted manner,’ explains Dr Carsten Marr, Director of the Institute of AI for Health at Helmholtz Munich.
His colleague Prof Julia Schnabel, Head of the Institute for Machine Learning in Biomedical Imaging and Professor of Computational Imaging and Artificial Intelligence in Medicine at TUM, sees DeepNeo as a building block for an AI-supported healthcare system that could support clinical decisions with unprecedented certainty in the future.
On the way to clinical application
The project was funded with a Helmholtz Innovation Grant and a patent application has already been filed. Ascenion, the technology transfer partner in the field of life sciences, is supporting the DeepNeo team in its search for potential industrial partners.
‘DeepNeo facilitates and standardises the evaluation of OCT imaging after stent implantation and thus contributes to more informed clinical decisions,’ say PD Dr Philipp Nicol and Prof. Dr Michael Joner, cardiologists at TUM Hospital, who provided clinical support for the project. ‘The procedure has the potential not only to reduce healthcare costs, but also to pave the way for more effective and personalised cardiovascular therapies.’
Original publication:
Deep learning model DeepNeo predicts neointimal tissue characterization using optical coherence tomography, Koch et al., Nature Communications Medicine 2025
Scientific contact:
Dr. Carsten Marr
Director of the Institute for AI in Healthcare, Helmholtz Centre Munich
E-Mail: carsten.marr(at)helmholtz-munich.de
Source: Press release of the Helmholtz Centre Munich
