Novel imaging biomarkers in age-related macular degeneration

Age-related macular degeneration (AMD) is the leading cause of blindness in the elderly. While the disease can be detected and monitored using optical coherence tomography (OCT) imaging, existing imaging biomarkers only allow for a coarse stratification of patients. To better treat the disease, there is a clear need for new imaging biomarkers that better describe patients’ physiology and provide improved risk stratification.

In the scope of the PINNACLE study, we have developed a data-driven approach to automatically propose candidates for novel AMD imaging biomarkers. It uses self-supervised deep learning to discover – without any clinical annotations – image features relating to both known and unknown AMD biomarkers. To interpret the learned biomarkers, we partition the images into 30 clusters that contain similar features. These clusters are reviewed and interpreted by teams of clinical experts.

Interactive AMD atlas

The resulting interactive AMD atlas is presented below. Each circle represents one of 46,496 OCT images, with colors corresponding to the 30 different clusters. Hovering over a circle displays the image as well as the cluster name that was provided by the clinical experts.

Related publications

For further reading, please refer to our two publications:

  • Holland et al. “Deep Learning–Based Clustering of OCT Images for Biomarker Discovery in Age-Related Macular Degeneration (PINNACLE Study Report 4).” Ophthalmology Science 4.6 (2024): 100543.
  • Holland et al. “Clustering disease trajectories in contrastive feature space for biomarker proposal in age-related macular degeneration.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer Nature Switzerland, 2023.
Link to the Ophthalmology Science paper Link to the MICCAI paper

The first paper covers the used dataset and deep-learning-based biomarker discovery strategy in great detail. Additionally, it describes the expert-in-the-loop setting for the interpretation of the discovered biomarkers and places our findings into clinical context. The second work extends our method to process time series of OCT images, so that the resulting clusters are able to capture the dynamic nature of AMD.

About the PINNACLE consortium

The PINNACLE consortium is an international, multi-disciplinary collaboration between researchers from Basel, Michigan, London, Southampton, and Vienna with expertise in ophthalmology, genetics, advanced image analysis, machine learning, and trial statistics. The consortium researches the pathophysiology of early- and intermediate-stage AMD in order to develop advanced machine learning algorithms to predict the conversion to advanced stages of the disease, and ultimately better treat affected patients. The PINNACLE study is supported by a Wellcome Trust Collaborative Award, “Deciphering AMD by deep phenotyping and machine learning” Ref. 210572/Z/18/Z.

Logo of the PINNACLE consortium