The Fraunhofer Heinrich Hertz Institute (HHI) has kicked off its work on the research project IToBoS "Intelligent Total Body Scanner for Early Detection of Melanoma".
The project's objective is to develop an AI-based diagnostic platform to advance the early detection of melanoma, a particularly aggressive type of skin cancer. IToBoS is scheduled to run from April 2021 to March 2025. The EU is funding the project with 12.1 million euros as part of the Horizon 2020 program. Out of this sum, Fraunhofer HHI will receive 810.000 euros for its research contribution.
The new platform allocates health data from various sources, including medical records, genomic data and in-vivo imaging. In addition, a full-body scanner and Computer Aided Diagnostics (CAD) tools will be connected to the platform. This creates a rich data pool for diagnoses utilizing AI models. With its explainable AI (XAI) methods, Fraunhofer HHI contributes to the visualization, explanation and interpretation of AI decisions made in the context of diagnosis. Moreover, the research team is incorporating methods of XAI already in the development process of the platform for ensuring the reliability of the AI models used.
Melanoma are responsible for 60 percent of all fatal skin neoplasms, which are malignant growth of the skin. The frequency of melanoma has increased significantly in recent years. They are often detected too late, as diagnostic procedures for early detection are labor-intensive and expensive.
To ensure early detection, a full-body skin examination is important. During this examination, which is the primary screening mechanism for melanoma, dermatologists individually examine each pigmented skin lesion looking for typical signs of melanoma. The IToBoS project aims to advance this examination method by employing an AI-based total body scanner. The scanner automatically examines the entire body within minutes using a cognitive AI assistant. As a result, it provides healthcare professionals with a risk assessment for each potential area.
For this task, the scanner is equipped with high-resolution cameras fitted with liquid lenses. These novel lenses, based on two immiscible liquids with different refractive indices, deliver unprecedented image quality. Using machine learning, these images are then integrated into the AI-based diagnostic platform with all the other available patient data. Thus, the platform supports dermatologists in providing a fast, reliable and highly personalized diagnosis of melanoma.
"An important task when utilizing AI in medicine is the continuous verification of the developed methods and data with XAI techniques," explains Dr. Wojciech Samek, head of the Department "Artificial Intelligence" at Fraunhofer HHI. "This prevents model misbehavior and ensures that our AI systems are reliable and safe."
To achieve this reliability, Fraunhofer HHI’s Department of “Artificial Intelligence” integrates novel XAI techniques into the employed AI tools. This allows to remove black-box limitations of current AI-powered CAD systems and enables making their decisions transparent. As another part of the project, the researchers collect and annotate a so-called ground truth dataset, which can form a basis for future analysis of computer vision data in this area.
Alongside Fraunhofer HHI, the IToBoS project consortium consists of Universitat de Girona , Optotune AG, IBM Israel - Science and Technology Ltd., Robert Bosch España SA, BARCO NV, National Technical University of Athens, Leibniz University Hannover, Fundació Clínic per a la Recerca Biomèdica, RICOH Spain IT Services SLU, Trilateral Research Limited, Università degli Studi di Trieste, Coronis Computing SL, Torus Actions SAS, V7 Ltd, Isahit, University of Queensland, Magyar Tudományos Akadémia Számít. és Automatizálási Kutatóintézet and Melanoma Patient Network Europe.
The article is also available on the Fraunhofer HHI website.