Google is one of the big players in the AI field, and here, its deep learning technology has proven that it is more accurate than pathologists at detecting breast cancer that has spread to a patient’s lymph nodes.
The AI detects the presence or absence of nodal metastases that influence a patient’s prognosis and treatment plan.
This makes it accurate in detecting breast cancer, helping the doctors detect other occurring symptoms.
Doctors can usually detect 38 percent of small metastases when samples are reviewed under time constraints. This pathologist’s examination has become the gold standard in diagnosis of nodal metastases.
The reason for Google's AI achievement, is because the machine learning technology called LYNA (LYmph Node Assistant) has been taught to learn about cancer by seeing many images of them, in a way similar to how a pathologist examines slides.
The algorithm’s first test showed that LYNA was able to correctly distinguish a slide with cancer from a slide without 99 percent of the time.
And because the AI was trained with slides in different magnifications, it was also able in detecting breast cancer in regions that were too small to be detected by pathologists.
According to a Google blog post:
"However, studies have shown that about 1 in 4 metastatic lymph node staging classifications would be changed upon second pathologic review, and detection sensitivity of small metastases on individual slides can be as low as 38% when reviewed under time constraints."
In the next test, six pathologists completed a diagnostic test with and without LYNA’s assistance.
With LYNA’s help, the doctors found it ‘easier’ to detect small metastases, and on average the task can take half as long.
Pathologists working with LYNA’s assistance were more accurate than both unassisted pathologists and the LYNA algorithm working alone.
Google’s researchers suggest that algorithms like LYNA could help with these identification tasks to allow more time for pathologists to work on more complex diagnoses.
"These studies have important limitations, such as limited dataset sizes and a simulated diagnostic workflow which examined only a single lymph node slide for every patient instead of the multiple slides that are common for a complete clinical case," continued Google.
The researchers want to test this algorithms further, to determine whether LYNA can work in real clinical workflows and patient outcomes, which involve a wider range of samples from different organs in the body.