Google is investigating how machine learning can be used to help with diagnosing breast cancer to ease pathologists' workload.
The tech giant's research team participated in the 2016 ISBI Camelyon Challenge, which sought to evaluate new and existing algorithms for automated detection of cancerous cells to reduce reading time for pathologists, and subjectivity in diagnosis.
The reviewing of pathology data is a complex task, requiring years of training to gain the expertise and experience to do well, says Google.
However, "even with this extensive training, there can be substantial variability in the diagnoses given by different pathologists for the same patient, which can lead to misdiagnoses", Martin Stumpe, technical lead, and Lily Peng, product manager, explain in a blog post.
The team provides the example of a study where pathologists were asked to independently diagnose a series of breast biopsies. It found agreement in diagnosis for some forms of breast cancer can be as low as 48%.
"The lack of agreement is not surprising given the massive amount of information that must be reviewed in order to make an accurate diagnosis," says the Google team.
The challenge focused on sentinel lymph nodes of breast cancer patients. A large dataset of images from the Radboud University Medical Centre and University Medical Centre Utrecht, both in the Netherlands, was provided to the participants.
Google used deep learning, a form of machine learning that can be applied to large datasets to help recognise patterns and images, to create an algorithm which goes through the pathology data faster than a human.
"We used the images to train algorithms that were optimised for localisation of breast cancer that has spread to lymph nodes adjacent to the breast," says the Google team.
"After additional customisation ... we showed that it was possible to train a model that either matched or exceeded the performance of a pathologist who had unlimited time to examine the slides."
It found the algorithm was good at identifying potential cancer but also identified a lot of false positives.
The team says: "These algorithms perform well for the tasks for which they are trained, but lack the breadth of knowledge and experience of human pathologists.
"To ensure the best clinical outcome for patients, these algorithms need to be incorporated in a way that complements the pathologist's workflow. We envision that an algorithm such as ours could improve the efficiency and consistency of pathologists."
Google says the research will not be incorporated into a real product yet as it will still have to go through clinical validation and regulatory approval. However, it hopes by sharing the work it will accelerate progress in the space.
The full report with technical details is available here.
Share