Towards integrated diagnostics in oncology: AI-driven predictive modeling of noninvasive multimodal data for cancer assessment and prognosis

Cancer is one of the leading causes of illness and death worldwide, and both early detection and accurate predictions about patient outcomes are essential for improving care. In daily practice, clinicians rely on multiple different tests, such as medical scans, blood tests, and clinical examinations, to understand the disease, choose treatments, and track progress. However, interpreting this information can be subjective, time-consuming, and variable between clinicians.
With hospitals now collecting large amounts of diverse medical data, this thesis explores how artificial intelligence (AI) can potentially support diagnosis and prognosis by combining routinely available, noninvasive tests. It evaluates different AI methods across multiple cancer types, treatments, diagnostic modalities, and predicted outcomes, showing where multimodal AI improves performance and where strong single tests already provide sufficient insight.
The thesis also incorporates follow-up tests collected during treatment to better reflect how the disease and treatment response evolve over time.