gms | German Medical Science

The potential of AI approaches approaches, such as Machine Learning (ML), cannot yet be fully assessed for bio-medical informatics. During this session, their limitations and opportunities will be presented in the domains of cancer treatment or *omics and discussed during a panel. Invited presentations will show international state-of-the-art research and corresponding application of A.I. in the domains cancer treatment, *omics data processing and imaging. This gives insights into the potential of ML-techniques with examples of clinical relevance.

• Dr. Okyaz Eminaga, Stanford University
  The first lecture addresses the application of artificial intelligence in the field of urology. Artificial neural networks can be used for the diagnostic classification of cystoscopic findings. After image preprocessing, three deep convolutional neural network (CNN) models were applied and evaluated showing correctly determined cancer lesions in all three models. Next steps will focus on integration of artificial intelligence-aided cystoscopy into clinical routines and possibly expansion to other clinical endoscopy applications. Deep learning models and machine learning approaches may also be useful instrument to to group patients with prostate cancer into clinically meaningful groups based on clinical parameters like Gleason score changes and tumor status changes.
• Dr. Sebastian Robert, Fraunhofer IOSB
  The presentation gives insights into the secure application of AI approaches in a Molecular Tumor Board (MTB) setting. Based on a digital twin patient model, we implement a medical data space to ensure a trustworthy infrastructure for the secure exchange of health data between distributed actors. The usage of the data can be controlled in a user-friendly, transparent, and data protection compliant manner. The analysis module is integrated in a trusted connector to guarantee data sovereignty of the patients. We use process mining techniques such as alpha-algorithm, heuristic, and inductive mining to identify treatment processes and visualize dependencies based on genetic alterations and clinical data. Finally, we adapt Graph Convolutional Neural Networks (GCNN) to facilitate graph process models for integrated clinical decision support and care research.
• Prof. Dr. Dennis Säring, FH Wedel
  In this talk a new ML based approach for the automated, computer-based, and non-invasive age estimation of male adolescents and young adults using three-dimensional (3D) magnetic resonance images (MRIs) of the knee is presented. The determination of certain age limits plays a crucial role in asylum applications, criminal proceedings, and professional youth sport whenever there is a lack of documentation or doubt about the alleged age. It can have important consequences for the persons in question, e.g. special benefits for underage refugees. The most common methods for age estimation in practice rely on the visual inspection of growth plate ossication of multiple long bones of the human body in X-ray or computed tomography (CT) images. However, the visual inspection is labour-intensive and subjective to the expert conducting the analysis. Moreover, the radiation exposure is considered “harm to the body” and should be used only as a last option. To overcome these disadvantages, non-invasive, automated, and unbiased methods for age estimation are required.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.