An algorithm associated with Google’s DeepMind is more successful at detecting breast cancer than experts in the field, according to a new study. The deep-learning initiative was able to spot breast-cancer in radiology scans either as accurately or better than radiologists.
Early stage
Although this research remains at an early stage, it is hoped that Google’s DeepMind will eventually be able to assist with the correction of misaligned diagnoses. Furthermore, such machine learning techniques could assist with the shortage of radiologists in some countries. It is envisaged that the algorithm can play a role in the early detection of breast cancer, which is key to treatment of the condition.
While all women aged over 50 in the United States and United Kingdom are currently screened for breast cancer, diagnoses can either be particularly distressing, or ultimately deadly.
Although this latest development is potentially excellent news for the treatment of breast cancer, the involvement of Google in healthcare systems remains controversial. Google has worked closely with NHS organizations in the UK, but some of the data that Google has acquired through this arrangement has been criticized.
Major breakthroughs
Nonetheless, there have been major breakthroughs achieved thanks to machine learning techniques, with this latest innovation suggesting that artificial intelligence will be able to read mammograms more accurately than the world’s top experts. And, naturally enough, representatives from Google Health were extremely enthusiastic about the results.
“This is another step along the way of trying to answer some of the questions that will be critical for us to actually deploying this in the real world,” Dominic King, director and UK lead of Google Health, commented in a statement. “This is another step closer to trying to deploy this type of technology safely and effectively.”
The Google algorithm was first used on 76,000 mammograms submitted by British women, via Cancer Research UK’s OPTIMAM dataset. This was supplemented by a further 15,000 scans from the United States. Once the machine learning system has been effectively trained, it was then tested on a further 28,000 scans from US and UK sources.
Google’s DeepMind assigns each mammogram a score based on three different modelling systems. The algorithm uses a neural network in order to analyse the information, with the conclusions of the system being compared against real-life results.
Real-world accuracy
Researchers claim that the AI model can predict breast cancer with the same level of accuracy as an expert radiologist. Indeed, the system was able to reduce false positives by 5.7% in the United States and 1.2% in the United Kingdom, and false negatives by 9.4% in the United States and 2.7% in the United Kingdom.
However, it should be noted that mammograms are usually checked by multiple radiologists in order to iron out any potential deficiencies in the system. Regardless of this, it seems that Google’s DeepMind could play a major role in breast cancer diagnosis in the future, and that this will be indicative of a growing role for machine learning in healthcare.
