Google Improves Artificial Intelligence's Ability to Detect Breast Cancer

2020-01-03

Two days ago, Google Health shared an artificial intelligence model that has been trained on over 90,000 x-ray images and achieved better results than human radiologists. It also identified more false negatives, which are images that appear normal but contain breast cancer. The initial findings were published in the journal Nature. The model not only achieved lower false positive rates (5.7% in the US and 1.2% in the UK), but also reduced false negative rates (9.4% in the US and 2.7% in the UK). According to data from the UK National Health Service and the American Cancer Society, approximately one in eight women are diagnosed with breast cancer at some point in their lives, making it the most common cancer diagnosis among women. The model was trained on breast x-ray images from 76,000 women in the UK and over 15,000 women in the US, and then evaluated using a dataset of over 25,000 images from UK women and over 3,000 images from US women. Google stated in a blog post, "Looking ahead, there are promising signs that this model could improve the accuracy and efficiency of screening programs, as well as reduce patient waiting times and stress. However, achieving this goal will require ongoing research, prospective clinical studies, and regulatory approval to understand and demonstrate how software systems inspired by this research can improve patient care." A year ago, there were reports that Google was developing artificial intelligence for detecting breast cancer. IBM's research has also focused on breast cancer. In 2019, IBM created a model that predicts when women will develop malignant breast cancer within a year and uses artificial intelligence to analyze breast cancer cells. Despite the numerous related studies, large-scale implementation has yet to be achieved, and the specific reasons for this are unknown. Recently, Google has once again improved the accuracy of breast cancer diagnosis, although their previous accuracy rate was already reported to be 99% (according to Google's own data). How much more accurate it has become is not specified. Laboratory products are often idealized, and when applied in reality, problems can arise due to the unpredictable nature of real-world scenarios. Perhaps this is why Google has made significant efforts to further improve accuracy. The 99% accuracy achieved in the lab may not necessarily translate to the same level of accuracy in real-world applications, leaving room for improvement for Google.

RECENT AI NEWS

RECENT AI TOOLS