GeoSOP: Geometry for surgical outcome prediction
Mentored by Stefanie Speidel & Guido Montufar
at National Center for Tumor Diseases (NCT), UKD Dresden and Max Planck Institute for Mathematics in the Sciences, Leipzig
Most surgeries are associated with postoperative complications. These complications can be life threatening or cause delays in recovery and further therapy, especially in the case of high-risk cancer patients. AI-based approaches for the prediction of post-operative complications can assist clinicians in surgical planning, post-operative patient management and treatment planning. It also encourages critical analysis of alternate surgical approaches and enables patient transparency. These AI-based approaches extract biomarkers from multi-modal data like CT/MRI scans, surgical videos, IR-spectral and flow cytometry data, and model a relationship between the extracted biomarkers and a specific postoperative complication.
Tapping the maximum potential of data is crucial to obtaining a reliable and accurate model. Previous studies that exploit the intersection of machine learning and geometry have demonstrated that the performance of machine learning models can be enhanced by processing real world data in its natural form without distorting its structural information. Therefore, in this PhD project, we aim to apply concepts of geometry to various phases of the complication prediction pipeline - (i) in analysing and organising high dimensional multi-modal data, (ii) in finding new biomarkers by utilising structural information in the data, (iii) in analysing patient populations and sub-groups, and (iii) in establishing causal structure in the data.
The main research questions in this context would be:
- How to apply geometry for representation learning of multi-modal data?
- How to build reliable machine learning models that can work well with irregular structured data? (Eg.: Graph Neural Networks)
- How to make the developed machine learning models transparent?
- How to find clinically relevant biomarkers for individualised outcome prediction?
- How to discover causal structures between the biomarkers and outcome by making use of geometrical deep learning techniques?
- How to translate the developed models into the clinic?
You will be working at the National Center for Tumor Diseases (NCT) Dresden at the medical faculty in collaboration with the Mathematical Machine Learning group at the Max-Planck-Institute of Mathematics in Sciences in Leipzig.
The NCT Dresden is located at the medical campus of TU Dresden and focuses on machine learning for cancer therapy by analysing data along the surgical treatment path to improve patient outcome.
The Mathematical Machine Learning group in Leipzig investigates geometry of neural networks and exploits innovative mathematics, drawing on information geometry and algebraic statistics.
- Master’s Degree (or equivalent) in computer science, mathematics, electrical engineering or related fields of expertise
- Very good programming skills (e.g. C++, Python)
- Excellent skills and practical experience in one or more of the following research areas is beneficial:
- Machine Learning
- Computer vision
- Ability to collaborate well in an interdisciplinary environment