By combining deterministic agent-based modeling, Bayesian inference, and foundational LLM models, it allows users to visualize cellular growth, test clinical treatments, generate synthetic biopsy slides, and receive real-time AI pathology reports.

 

⚠️ Disclaimer: Limitations & Potential Biases

​

Strictly for Educational Use: OncoTwin is a conceptual demonstration and educational tool. It is not an FDA-approved medical device and must never be used for real-world clinical diagnosis, prognosis, or treatment planning.

Biological Simplification: The underlying physics engine utilizes a 2D cellular automaton and a simplified Bayesian network. It fundamentally abstracts the extreme complexity of in vivo tumor microenvironments, 3D spatial heterogeneity, systemic immune responses, and multi-omics data.

AI Hallucinations & Training Bias: * Generative Text: The clinical reports and "Tumor Twin" chatbot are powered by Large Language Models (LLMs). These models predict statistically likely text based on training data and may hallucinate medical facts or invent false clinical correlations.

​

• Synthetic Media: The generated H&E biopsy slides are illustrative synthetic artifacts, not scientifically accurate representations.

• Dataset Bias: The AI models may exhibit biases inherent to their training datasets, potentially over-representing common oncological presentations (typically from Western/Caucasian demographics) while under-representing rare pathologies or diverse patient populations.