https://www.nature.com/articles/s41591-023-02700-1
Six ways large language models are changing healthcare
…
Predicting cancer metastasis
LLMs are being used to predict metastatic cancer and to assist in devising clinical treatment responses, says Amber Simpson, Canada Research Chair in Biomedical Computing and Informatics at Queens University, Canada, who collaborates with a research team at Memorial Sloan Kettering Cancer Center in New York City.
Her research on LLM-assisted cancer diagnosis includes a set of studies in which cancer-progression patterns were extracted from vast numbers of computed tomography reports (714,454 structured radiology reports were used in the most recent study2) and were analyzed by a natural language processing model that was then used to predict metastatic disease in multiple organs. The models used features from consecutive structured text radiology reports to predict the presence of metastatic disease.
By offering oncologists a predictive pathway of how cancer will progress, the LLM shifts treatment strategies toward a more targeted, deliberate approach.
“Currently, we treat everyone to help a few,” Simpson explains. “With this type of model, if you add the treatment data, you can start saying the next treatment this patient should receive is this specific therapy — and that’s really the holy grail of precision medicine.”
Such an approach helps to elucidate cancer drugs beyond clinical trials, in real-world settings, Simpson adds. “This type of analysis allows us to understand population-level analysis of cancer patients and how drugs work in the wild.”
Unusually for an AI model, this is already in clinical use. “We’re already showing this approach can be used clinically. We’re already targeting patients,” says Simpson.
…