The war on cancer has gained a powerful new ally in the form of neural network technology that can analyze medical images and identify cancerous tumors with greater accuracy than human pathologists. Google announced its GoogLeNet AI object recognition technology successfully detected malignant tumors in breast cancer images 89 percent of the time, compared to an average of 73 percent for human experts.
GoogLeNet employs deep learning technology to perform digital pathology. The company said it created an automated detection algorithm to examine breast cancer images. The algorithm was specifically designed to localize breast cancer that has spread to lymph nodes near the breast.
To train the algorithm, Google used images from the Radbound University Medical Center in The Netherlands.
Google’s technology addresses major challenges involved in diagnosing and prescribing therapy for cancer patients. Pathologists require years of training to learn cancer detection techniques. And even after extensive education, the process of pathology is highly subjective, with professionals agreeing on a diagnosis less than half the time for some forms of breast cancer, according to Google.
This major uncertainty factor reflects the huge quantity of information that must be analyzed during a cancer diagnosis. A single patient’s condition is often displayed in multiple slides, each depicting an image of more than 10 gigapixels. Experts often have limited time to view all this information, leading to further variability in diagnoses.
Google noted that its technology worked well for other information beyond the training data set, including using images from a different hospital using different types of scanners. The company detailed its findings in a report entitled, “Detecting Cancer Metastases on Gigapixel Pathology Images”.
Stephan Cunningham is vice president, product management at Veritone. Working in concert with core internal teams including industry-specific general managers and engineering as well as directly with clients and prospects, he leads the disciplines and business processes which govern the Veritone Platform.