In its quest to make AI more useful, Google has come up with a new algorithm for doctors. The Google AI tool will assist doctors in detecting breast cancer by identifying metastatic (advanced) breast cancer more accurately than pathologists can by looking at slides.
How can it help in detecting breast cancer?
The new Google AI tool is called the LYmph Node Assistant, or LYNA. The tool will work sort of like a “spell check” for doctors responsible for diagnosing cancer by examining images of a patient’s cells. Lymph node tissues help in detecting whether a patient’s breast cancer has spread to other parts of their body. Pathologists study the tissue samples from breast cancer patients’ lymph nodes to confirm how much the tumor has spread and how dangerous it can be.
Metastatic tumors are cancerous cells which break away for their tissue of origin and travel through the body to form new tumors in other parts of the body. Such tumors are known to be difficult to detect. A 2009 study of 102 breast cancer patients at two Boston health centers revealed that one in four suffered from “process of care” failures like incomplete diagnostic tests and inadequate physical examinations.
This is probably a major reason that an estimated 90% of the 500,000 deaths worldwide caused by breast cancer can be attributed to metastasis.
Google AI tool is better than humans, but not perfect
Last week Google released two research papers, which were published in The American Journal of Surgical Pathology and the Archives of Pathology and Laboratory Medicine. The first paper detailed how the tool can assist in identifying cancer cells on tissue images.
Google’s second paper is about how the tool worked with pathologists and also individually. The papers state that the Google AI tool has an accuracy rate of 99% when it comes to identifying which slides show metastatic cancer. In comparison, pathologists reportedly miss metastases on individual slides 62% of the time when operating under time limits.
“[LYNA] achieves higher tumor-level sensitivity than, and comparable slide- level performance to, pathologists,” the researchers wrote. “These techniques may improve the pathologist’s productivity and reduce the number of false negatives associated with morphologic detection of tumor cells.”
However, Google’s LYNA isn’t perfect. On a few occasions, it misidentified giant cells, germinal cancers, and bone marrow-derived white blood cells known as histiocytes.
Google stressed that it is important for pathologists and the algorithm to work together. The results were remarkable when LYNA worked as a companion to pathologists. Using the Google AI tool not only lowered the rate of missed micro-metastases by a “factor of two,” but it also brought down inspection times considerably.
“Artificial intelligence algorithms can exhaustively evaluate every tissue patch on a slide,” Google’s paper read. “We provide a framework to aid practicing pathologists in assessing such algorithms for adoption into their workflow (akin to how a pathologist assesses immunohistochemistry results).”
Google’s AI tool LYNA is based on Inception-v3, an open-source image-recognition deep-learning model. This deep learning model achieved a 78.1% accuracy rate on Stanford’s ImageNet data set. The Google AI tool used a 299-pixel image as input, highlighted tumors at the pixel level and extracted labels of the tissue patch.
To perfect this Google AI tool, the search giant used a de-identified data set of breast cancer patients’ lymph node scans from medical centers in the Netherlands. The team used two sets of pathological slides to train the algorithm to identify characteristics of tumors in varying conditions.
Even though the Google AI tool is yet to be used in real-life clinical situations, doctors are acknowledging its usefulness and how its role can be expanded. When it is ready for practical use, it could lead to more accurate diagnoses and free up doctors to focus more on treating patients.
Google is investing in health care applications for artificial intelligence. Recently, Google’s Medical Brain team claimed to have developed an AI system which could predict mortality rates with 90% accuracy. Additionally, Verily, which is Alphabet’s life sciences subsidiary, is developing a system that determines a person’s risk of heart disease using retinal scans.