Facilitating precision oncology with agentic AI based on clinical practice guidelines
ABSTRACT
Cancer care is becoming increasingly complicated as the number of cancer subtypes grows and the arsenal of treatment options expands, driving the rapid proliferation of clinical practice guidelines. For example, the European Society for Medical Oncology has now published more than 200 onoclogy decision trees covering treatment of different cancer types, tumour subtypes, and disease stages. While valuable, such resources impose a tremendous cognitive burden on oncologists, especially given the rapid pace of treatment advancements and guideline updates. Agentic AI could help address this challenge, provided that the system is equipped with robust guardrails to ensure determinism, correctness, and safety.
Research objectives:
1. To develop decision trees rooted in clinical practice guidelines.
2. To develop an attribute harmonization system for cancer patient data.
3. To implement a decision-support system based on agentic decision trees.
4. To establish realistic benchmarks for oncological decision making.
5. To evaluate and benchmark agentic decision trees against conventional AI agents.
SKILLS & EXPERIENCE REQUIRED
- Major in Computer Science, Artificial Intelligence, Biomedical Sciences, Biomedical Engineering, or related disciplines
- Solid foundation in pure or applied mathematics
- Experience using Git and applying best practices of software development
- Proficiency in Python without GenAI assistance
- Attention to detail