Google LLC’s health unit has hit the ground running in 2020, today detailing a homegrown artificial intelligence model that it says can identify breast cancer more accurately than human radiologists.
The project is the fruit of transatlantic research collaboration. Google Health worked together with Cancer Research UK Imperial Centre, Northwestern University, Royal Surrey County Hospital and DeepMind, Google parent Alphabet Inc.’s AI development unit.
The search giant taught the machine learning model to identify cancerous tissue by training it on anonymized mammograms from more than 91,000 women in the U.S. and the U.K. Then, Google put the neural network to the test by having it analyze a separate dataset of scans from 28,000 patients.
The model managed to detect tumors with a lower error rate than a panel of six radiologists who took part in the study. When evaluating scans taken in the U.S., Google’s AI produced 5.7% fewer false positives than the human team and recorded 9.4% fewer false negatives, or instances when cancerous growths are mistaken for healthy tissue. The model beat the experts in the U.K. as well with 1.2% fewer false positives and 2.7% fewer false negatives, respectively.
“Notably, when making its decisions, the model received less information than human experts did,” Google Health researchers Shravya Shetty and Daniel Tse wrote in a blog post.
“The human experts (in line with routine practice) had access to patient histories and prior mammograms, while the model only processed the most recent anonymized mammogram with no extra information. Despite working from these X-ray images alone, the model surpassed individual experts in accurately identifying breast cancer.”
Tackling breast cancer is a natural area for Google to focus on its healthcare AI efforts.
According to statistics from the American Cancer Society, about one of every 8 women in the U.S. will develop the disease over her lifetime. But catching tumors early is difficult because manually evaluating mammography images is technically challenging and some countries have a shortage of radiologists who can perform the task.
Google believes that its AI has the potential to ease early detection as well as reduce wait times for patients by freeing up human medical staff’s time.
The company has already explored ways of applying technology in the real world. During the study, Google ran a simulation that saw the AI participate in a double reading, a process wherein a mammogram is evaluated by two radiologists to increase detection accuracy.
Google’s model played the part of the first reader and reduced the workload of the second expert by 88%, according to the search giant’s Nature paper on the study.
The initiative is the latest in a series of promising healthcare AI projects from the Alphabet family of companies. Previously, Google demonstrated a model that can detect lung cancer in CT scans and sister firm Verily showed a method of using retinal images to identify risk factors for cardiovascular disease.