Personalized medicine focuses on customizing treatments for each patient. So far, this has typically relied on a limited set of factors to forecast disease development. However, these limited factors often fall short of capturing the intricate nature of diseases like cancer. A research team has introduced a novel solution to this challenge with the help of artificial intelligence (AI).
Personalized medicine focuses on customizing treatments for each patient. So far, this has typically relied on a limited set of factors to forecast disease development. However, these limited factors often fall short of capturing the intricate nature of diseases like cancer. A research team from the Faculty of Medicine at the University of Duisburg-Essen (UDE), LMU Munich, and the Berlin Institute for the Foundations of Learning and Data (BIFOLD) at TU Berlin has introduced a novel solution to this challenge with the help of artificial intelligence (AI).
Utilizing the advanced medical infrastructure at University Hospital Essen, the researchers have combined various data sources—such as medical histories, lab results, imaging, and genetic analyses—to enhance clinical decision-making. “Despite the vast amounts of clinical data in modern healthcare, the promise of genuinely personalized medicine often remains unmet,” notes Prof. Jens Kleesiek from the Institute for Artificial Intelligence in Medicine (IKIM) at University Hospital Essen and the Cancer Research Center Cologne Essen (CCCE). Current oncological practices often rely on rigid evaluation systems, like cancer stage classifications, which overlook personal differences such as gender, nutritional health, or existing conditions. “Modern AI technologies, especially explainable artificial intelligence (xAI), can analyze these complex relationships and significantly enhance the personalization of cancer treatment,” states Prof. Frederick Klauschen, Director of the Institute of Pathology at LMU and research group leader at BIFOLD, where this approach was developed alongside Prof. Klaus-Robert Müller.
In a new study published in Nature Cancer, the AI system was trained on data from over 15,000 patients with 38 types of solid tumors. The study explored how 350 different parameters interact, including clinical details, lab results, imaging data, and genetic tumor information. “We identified crucial factors that drive most of the decision-making in the neural network, along with numerous significant interactions between these parameters,” explains Dr. Julius Keyl, Clinician Scientist at the Institute for Artificial Intelligence in Medicine (IKIM).
The AI model was subsequently validated using data from more than 3,000 patients with lung cancer to confirm the identified interactions. By synthesizing this data, the AI generates a comprehensive prognosis for each patient. As an explainable AI, the model provides transparency for clinicians by illustrating how each factor influenced the prognosis. “Our findings showcase AI’s ability to analyze clinical data in context rather than in isolation, allowing for a re-evaluation and enabling personalized, data-driven cancer treatment,” says Dr. Philipp Keyl from LMU. Such AI methodologies could also be applicable in emergency situations where rapid assessment of diagnostic parameters in their entirety is crucial. The researchers also plan to explore complex interactions across different types of cancers, which conventional statistical methods have overlooked. “At the National Center for Tumor Diseases (NCT), in collaboration with other oncological networks like the Bavarian Center for Cancer Research (BZKF), we have the ideal environment to advance our work: demonstrating the tangible benefits of our technology in clinical trials,” adds Prof. Martin Schuler, Managing Director of the NCT West site and Head of the Department of Medical Oncology at University Hospital Essen.
