Digital twin for cancer treatment
Mapping a venture idea on potential against difficulty
/images/digital-twin-cancer/hero.webp
Value proposition board listing the venture's core abilities, an LLM interpreting patient notes and medical literature, and market opportunities across cancer treatment, R&D and education
- Year
- 2024
- Role
- Sole author
- Tools
- Opportunity assessment · Regulatory barrier analysis
- Contribution
- Sole author, individual opportunity assessment
An opportunity assessment of a venture concept: a digital twin of cancerous nodules running with a large language model, to predict tumour growth and treatment response. Both the potential and the difficulty came out high.
The idea
A digital twin of cancerous nodules, driven by a large language model, using scans, vitals and patient history together with medical literature to predict how a tumour grows, how it responds to treatment, and how drug resistance develops. It is meant to assist the doctor and reduce their workload, not to replace them.
Why it is a hard case rather than an easy one
The potential is high. There is a compelling reason to buy, the market is very large, and margins are strong, particularly in a system like the US where hospitals will invest in better outcomes. But the difficulty is just as high. Development needs access to a lot of patient data and a twin that can actually interact with a language model, which could take a large team several years. Time to revenue could look like drug approval, which can run past ten years. And adoption has to get through regulation that differs by country, institutional approval, and real scepticism about AI making clinical decisions.
The part I would defend
The mitigations are where the thinking is. Scope it down so it assists doctors instead of replacing them, because that is much faster to get in front of people, especially in the private sector. Enter first where the regulatory barrier is lowest rather than where the market is biggest. And build trust with retrospective case studies, running the model against historical cases and comparing what it recommends to what the treating doctors actually did, so there is accuracy data before anyone is asked to rely on it.