GoogLeNet: Google Turns to Artificial Intelligence to Diagnose Cancer

Pathologists may finally have a partner in crime when it comes to determining whether or not a patient has cancer. Scientists are now collaborating with Google to create an artificial intelligence that can help pathologists detect and size up tumors for prognosis, and the results are promising.

According to Google's paper entitled "Detecting Cancer Metastases on Gigapixel Pathology Images," experts at the company are using GoogLeNet technology for the new cancer diagnosis system. Google apparently has been modifying its networks to analyze images the way pathologists do in order to have a model that can replicate the functions of pathologists, assuming they had all the time to analyze specimens.

The results were astonishing, as the new Google AI system was able to score a smooth 89 percent accuracy versus the 73 percent of human pathologists.

According to CNN, Lily Peng of Google explained how their new AI system was able to assess and detect potential breast cancer patients. However, Jeroen van der Laak from Radboud University Medical Center said that there are still improvements that are needed to be made in order to create a "good enough" algorithm to detect cancer. The AI system is capable of generating false reports, as there are certain factors in observing these photos that only trained pathologists can interpret correctly.

In a report from IBTimes, experts explained that much as pathologists are well-equipped to analyze patient data, there can be too much for them to handle.

For instance, they said that a patient can have a lot of slides to be observed, with each at 10 gigapixels when digitized at 40 times magnification. A pathologist may have to go through thousands of these photos just to make sure their data is accurate.

Interestingly, the applications of the system are impeccable. For instance, pathologists may be able to find warnings from the AI about certain tumor red zones and even accurately measure them for cancer prognosis.