Researchers have taught computer to analyze lung cancer tissue and distinguish between the two most common types of the disease — adenocarcinoma and squamous cell — with an accuracy of more than 90 percent.
RESEARCHERS HAVE TAUGHT A computer to analyze lung cancer tissue and distinguish between the two most common types of the disease — adenocarcinoma and squamous cell — with an accuracy of more than 90 percent.
Once the computer started learning to recognize the tissue, it was able to refine its process in ways even the researchers don’t fully understand, reported Cancer Currents, a blog run by the National Cancer Institute. Using a process developed and published by Google, the machine applied artificial intelligence to teach itself how to improve at making the diagnoses.
The study was conducted by researchers at New York University (NYU) School of Medicine, with results published Sept. 17 in the journal Nature Medicine. Ultimately, the computer matched pathologists’ accuracy in distinguishing between the two cancer types, but it completed the task in seconds rather than the minutes a doctor needs.
Furthermore, the machine fairly accurately predicted the presence of six genetic mutations that might be present in the tissue samples just by looking at them; humans must send tissue for DNA sequencing to get that information, with results taking up to two weeks, Cancer Currents reported. The machine’s predictions, according to the blog article, could help doctors choose the right medications for patients at the outset of treatment, rather than guessing and then potentially switching treatments when DNA results come back. The computer could also be used during surgery to confirm that surgeons have collected a usable biopsy and provide second opinions after pathologists analyze tissue.
“These findings suggest that deeplearning models can assist pathologists in the detection of cancer subtype or gene mutations,” the study authors wrote. “Our approach can be applied to any cancer type.”
At NYU, researchers are already using the code to teach machines to diagnose kidney, breast and other cancers, according to the blog.
In addition, researchers at the University of Pittsburgh are using artificial intelligence to analyze suspicious mammogram results and help determine whether the cases are malignant or benign before calling women back into the clinic for further testing.