Background

The demographic change of the aging population and the accompanying increasing cancer incidence poses a substantial threat to our healthcare system. Simultaneously, rapid advances in cancer diagnosis and treatment have made personalized health (PH) approaches available for some cancer types e.g. melanoma and lung cancer. Personalized health research is currently focused on high throughput analysis of biological specimens e.g. synthesis of deoxynucleic guanidine (DNG) by whole genome sequencing. In parallel, recent research aims at making “traditional” diagnostics such as radiological imaging and histopathological imaging available for deep quantitative analyses and incorporate numerous and diverse patient and disease characteristics into prognostic and predictive PH models. However, the quantitative amount and qualitative diversity of these characteristics poses a challenge for traditional outcome modelling, treatment stratification and response evaluation.

Within the last years, enormous progress has been achieved regarding artificial intelligence and machine learning. However, the focus of research has shifted. Algorithms such as deep neural networks, which have been known since many years, have gained new interest due to the recent availability of dedicated hardware (mostly graphics processing units with many parallel computing cores) at relatively low costs. It has become clear that not the knowledge of the algorithms or the required computer hardware themselves are valuable, but the availability of highly structured databases, which can be used to train machine learning programs. For this reason, the large companies (such as Google and Microsoft) and universities (Berkeley) have made their machine-learning libraries (Tensorflow, Caffe, Theano, Scikit-Learn) as open-source software freely available. The open release of these software packages resulted in a boost of medical imaging post-processing, which will fundamentally change the working environment of physicians within the next decade. The bottleneck now for the application of artificial intelligence in medicine is the expert knowledge of physicians regarding the relevant questions to be addressed, required medical input to train the algorithms and integration of artificial intelligence algorithms into human expert systems. This project aims to develop an artificial intelligence platform ultimately leading to clinical decision support systems to address the challenges in oncology.