Machine Learning Research

• Preprocessing
• Registration
• Segmentation
• Spatio-Temporal Modelling

Machine Learning

Spation-Temporal Modelling

Biomarkers derived from medical imaging data are an essential tool for diagnosis, therapeutic decisions, and evaluation of treatment response. They provide valuable insight by quantifying informative changes in anatomic, physiological, biochemical or molecular processes. In a clinical setting, predictive biomarkers that estimate future disease development and treatment response are exceptionally beneficial, since they allow to personalize treatment, and to optimize its effect.

Prediction of Retinal Disease Recurrence

We developed a method to predict treatment response patterns based on spatio-temporal disease signatures extracted from longitudinal SD-OCT images. These signatures describe the underlying retinal structure and pathology based on total retinal thickness maps.  The figure above illustrates 12 month follow-up series of aligned total retinal thickness maps for two patients with recurring edema at month 5 and 10 respectively without recurring edema. Based on morphology observed in the early stage of treatment, the algorithm predicts if edema will recur in an individual patient after treatment.

 “Spatio-Temporal Signatures to Predict Retinal Disease Recurrence.”
Wolf-Dieter Vogl, Sebastian M. Waldstein, Bianca S. Gerendas, Christian Simader, Ana-Maria Glodan, Dominika Podkowinski, UrsulaSchmidt-Erfurth, and Georg Langs.
In Proc. IPMI, 2015.