The last few years have seen a number of applications of AI in healthcare, and one of the more interesting is in predicting the outcome of particular treatments. The latest project of this ilk comes from researchers at Georgia Tech, who have developed a new tool to predict the outcome of chemotherapy.
The tool, which was documented in a recently published paper, is designed to analyze the specific RNA expressions that are tied to specific outcomes tied to chemotherapy drugs. When the tool was applied to real cases, it was able to present the best outcome around 80% of the time. The team believe there tool could prove invaluable, as whilst the first drug used to treat cancer is usually pretty straightforward, if that fails for whatever reason, the plan B treatment is much less clear cut.
“By looking at RNA expression in tumors, we believe we can predict with high accuracy which patients are likely to respond to a particular drug,” the researchers say. “This information could be used, along with other factors, to support the decisions clinicians must make regarding chemotherapy treatment.”
Precision medicine
The system was trained on detailed records of RNA from tumors alongside the outcome of treatments for specific drugs. Unfortunately, this isn’t a huge dataset, with just over 150 records available, so the team trained the system on just over 100 of these, before testing it on the remainder.
The team began their work by looking specifically at ovarian cancer, but rapidly expanded their scope to include a range of other cancer types, including lunh and breast cancer. They could do this because the same chemotherapy drugs are used for each form.
“Our model is predicting based on the drug and looking across all the patients who were treated with that drug regardless of cancer type,” the researchers explain.
The system aims to compare the likelihood that a particular drug will have an impact on the cancer of each patient, which will hopefully allow doctors to be better informed when determining the treatment given.
The team plan to make the system available as open source software to encourage as many people as possible to not only use it but to improve upon it. The team believe this is definitely possible, not least because it’s accuracy will improve as more data is analyzed by the system.
“To really get this into clinical practice, we think we’ve got to open it up so that other people can try it, modify if they want to, and demonstrate its value in real-world situations,” the team explain. “We are trying to create a different paradigm for cancer therapy using the kind of open source strategy used in internet technology.”
