- Researchers at Google have developed a deep learning tool that can identify metastasized breast cancer with 99 percent accuracy and could reduce the time it takes for clinicians to review pathology slides.
Detecting cancer that has spread from the primary site to nearby lymph nodes is a laborious process, particularly in the case of breast cancer. For this condition, nodal metastasis influences decisions associated with chemotherapy and radiation, making accurate and timely detection essential.
However, human pathologists often struggle to correctly identify the spread of the disease. Researchers cited previous studies that show about one in four metastatic lymph node classifications would be changed upon a second pathologic review, and that sensitive detection of small metastases on individual slides can be as low as 38 percent.
To enhance the identification process, the research team created the LYmph Node Assistant (LYNA) in 2017. The group tested the deep learning tool in two separate studies to ensure the algorithm would improve pathologists’ workflows and diagnostic accuracy.
In the first study, the team applied LYNA to de-identified pathology slides from two different datasets. Researchers found that with both datasets, the tool was able to correctly distinguish a slide with metastatic cancer from a slide without cancer 99 percent of the time.
Additionally, LYNA was able to accurately determine the location of both cancers and other suspicious regions within each slide, some of which were too small to be consistently identified by pathologists.
In the second study, researchers conducted a simulated diagnostic task in which they reviewed lymph nodes for metastatic cancer both with and without the assistance of LYNA.
When detecting small metastases, the team reported that using LYNA made the complex task objectively easier and halved average review time, requiring one minute instead of two minutes per slide.
“This suggests the intriguing potential for assistive technologies such as LYNA to reduce the burden of repetitive identification tasks and to allow more time and energy for pathologists to focus on other, more challenging clinical and diagnostic tasks,” said Martin Stumpe, Technical Lead and Craig Mermel, Product Manager, Healthcare, Google AI.
This research builds on a previous study from Google, in which investigators used deep learning and EHR data to predict mortality and hospital readmissions. The deep learning tool outperformed traditional prediction models in both speed and accuracy.
With this new venture, Google hopes to expand the use of deep learning into cancer care as well.
The team did acknowledge that their study had some limitations, namely limited dataset sizes and the use of a simulated diagnostic workflow. Further work will need to assess LYNA’s impact on real clinical workflows and patient outcomes.
Still, despite these limitations, the research team expects that tools developed in the future will improve diagnostic accuracy in healthcare.
“We remain optimistic that carefully validated deep learning technologies and well-designed clinical tools can help improve both the accuracy and availability of pathologic diagnosis around the world,” said Stumpe.