- Steinbeis Transfer-Magazin - https://transfermagazin.steinbeis.de -

AI for the Next Generation of Tumor Diagnostics

A research network develops a platform based on AI technology to support medical treatment

Genomic cancer medicine and immunotherapy are revolutionizing the treatment of cancer patients. U.S. scientists Ralph Weissleder and Mikael Pittet predict that in the future, most diagnostic and therapeutic decisions in oncology will be based on the molecular genetic analysis of tumor patients as part of precision medicine. Many new forms of cancer will likely be treated as chronic conditions as a result of these new approaches. In the future, tumors will be treated not only according to their location in the body, but also according to their genomic profile. This makes it necessary to analyze the genomic tumor profile of each patient and determine individual treatments on a case-by-case basis following reviews by molecular tumor boards. Steinbeis entrepreneur Professor Dr. Dirk Hempel and his Steinbeis Transfer Institute for Clinical Hematology-Oncology are part of a network that is working on the development of an AI-based platform to be used by virtual molecular tumor boards. The platform will allow experts to match genomic data with imaging and clinical courses to support decisions regarding appropriate treatment.

Personalized medicine is developing at such a rapid pace that it will soon find its way into routine healthcare, so it will have to be widely available. It will also not take long for diagnostics to reach such a high level that it will no longer be possible for a small number of leading specialist centers to deal with demand. “Since the expert teams who take part in molecular tumor boards are not available at all hospitals, our aim is to make an AI-based expert system widely available so it supports the virtual, web-based work of molecular tumor boards – so-called Virtual Molecular Tumor Boards,” says Dirk Hempel, outlining one of the network’s goals.

The challenge is that genomic tumor analysis involves generating and analyzing extremely large amounts of data (big data). In addition, individual patients’ treatable gene modifications come in a huge range of variants, and the incidence of driver mutations that can provide a point of attack for subsequent drug interventions is therefore very low.

The goals of the OnkoVision project

For the OnkoVision initiative, a team of researchers from the Steinbeis Transfer Institute for Clinical Hematology-Oncology has been working with the Fraunhofer Institute of Optronics in Karlsruhe (IOSB), the Helmholtz Institute in Munich, and the Technical University of Munich to develop an automated high-tech support platform using artificial intelligence. The planned platform differs fundamentally from all systems currently available in Europe, as well as international markets.

The network has defined four key questions that need to be addressed by the OnkoVision project:

  1. How does one achieve broad-scale genomic medicine?
  2. How can the enormous volumes of data be evaluated and compared with large databases?
  3. How should molecular tumor boards be implemented nationwide?
  4. How can the vast amount of real-world data be used for oncological healthcare provision research?

The role of OnkoVision

OnkoVision is looking into ways to develop and test AI solutions that will make it possible to link genomics repositories, including omics and medical databases, biobanks, and other registries, with the goal of supporting clinical research and decision-making. The project aims to combine the following functions:

The platform is being developed to include two modules. Module A will be used for the first step to facilitate decisions made by the molecular tumor board with the support of AI algorithms. The platform should make molecular medicine expertise available to a wider audience in order to facilitate molecular medicine as a basis for precision medicine – not just at a small number of leading centers, but also in rural areas. By making full use of AI, the system is expected to be self-learning and improve through continuous input from medical experts. Another role played by the system will be to tap into growing volumes of data by using machine learning to draw on so-called real-world data and answer research issues as part of medical research.

The second module should make it possible to detect new molecular biomarkers for use in diagnosis and treatment, such that eventually orphan drugs will be approved for rare tumor diseases based on real-world data.

The approach adopted for the platform reflects a new generation of decision-making support and differs significantly from all previously available systems. Of course, one essential factor for making optimal use of the system is that it is “networkable” – this means it should not only forge networks with other databases, but also network with users. The community of users should comprise certified tumor centers working in the clinical field as well as centers dealing with outpatients in the broadest possible sense – i.e. not just a small number of leading centers.