Researchers at the University Hospital of Cologne have developed a novel, agent-based AI system that fundamentally changes pathology. The framework SPARK (System of Pathology Agents for Research and Knowledge), designed as a “digital brain”, can extract previously hidden biological information from routine histological tissue sections and thus enable a much deeper understanding of tumor diseases.

In contrast to classic AI approaches, which are mainly limited to tissue segmentation or single-cell analyses, SPARK works autonomously. It combines several specialized algorithms into a coordinated system that independently generates biological hypotheses, refines them and translates them into concrete analytical tools – without retraining the models. The interaction takes place via natural language, so that even clinicians without programming knowledge can carry out complex evaluations.

In extensive validation studies with over 5,400 patients from 18 independent cohorts and five different tumor entities, SPARK proved to be highly accurate. The system identified clinically relevant tissue markers that correlate closely with disease progression, established pathological parameters and response to therapies. In addition, it makes it possible to draw conclusions about the temporal development of tumors from static sections.

The procedure helps to make diagnoses more precise, to better stratify patients and to make treatment decisions in a more targeted manner. Personalized oncology in particular opens up the possibility of adapting treatments more closely to the individual biological characteristics of a tumor.

Another advantage is the high accessibility thanks to an intuitive, modular user interface. All developed methods, parameters and results have been made openly available to promote further development by the scientific community.

The study, entitled “An agentic framework for autonomous scientific discovery in cancer pathology”, has been published in the journal Nature Medicine . The work was funded by the former Federal Ministry of Education and Research (now: BMFTR) and the European Interreg project DigiPathConnect, among others.

Original Paper:

An agentic framework for autonomous scientific discovery in cancer pathology | Nature Medicine

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability ected.