CD Laboratory for Machine Learning Driven Precision Imaging

This CD Laboratory will integrate radiological and pathological images as well as molecular data using new machine learning methods. This will lead to new predictive models for lung cancer and its individualised treatment. This will pave the way for the development of new machine learning (ML) concepts in precision imaging for better individualised treatments.

In order to be able to fight primary lung cancer (i.e. originating from the cells of the lung itself), which is one of the most common types of cancer and the most common cause of cancer death worldwide, more effectively in general and treat it more efficiently on an individual basis, the two co-heads Georg Langs and Helmut Prosch have designed the CD Laboratory for Machine Learning for Precision Imaging as an interdisciplinary project in which renowned scientists from the fields of machine learning, Medicine, oncology and pathology work together: Supported by team members dealing with practical legal issues and external experts contributing patient cohorts and expertise, the aim is to develop and validate a novel ML methodology that will be widely applicable in the future to improve the individualised care of lung cancer patients.

This very promising integration of AI (artificial intelligence) and imaging does, of course, present numerous challenges: Available routine clinical data is heterogeneous, the patient population is diverse, training data that can be used for AI models is limited in number and it is a highly complex endeavour to keep said models constantly up to date so that they work with the availability of new therapies and the simultaneous development of imaging technologies, while the legal conditions for sharing data collections are also a complicated issue.

In order to overcome these challenges, the overarching goal is divided into four sub-goals: Firstly, to quantitatively assess and predict disease and therapy progression; secondly, to extend machine learning to the large, diverse, heterogeneous routine patient population and to continuously evolving and emerging diagnostic technologies and treatment options, rather than simply starting from focussed studies. Thirdly, the aim is to use ML to link evidence in the form of large-scale data on the one hand with an understanding of underlying biological processes on the other, and finally, fourthly, the legal requirements for shared data use still need to be clarified, while AI models also need to be developed and trained on large medical data sets.

The future results of this ambitious project will make important contributions to improving the success of individual therapy and increasing the number of patients benefiting from ML in therapy, while at the same time creating a basis for methods whose applicability goes beyond lung cancer.