New breakthrough! Google’s breast cancer AI detection system misdiagnosed only 5.7% lower than humans

New breakthrough! Google’s breast cancer AI detection system misdiagnosed only 5.7% lower than humans

As we all know, in order to detect breast cancer as early as possible, many countries have implemented large-scale breast screening programs. On January 1, 2020, Google Health teamed up with artificial intelligence company DeepMind to release an artificial intelligence breast cancer detection system in the top academic journal Nature. The authors say the system’s ability to detect breast cancer exceeds that of professional radiologists and may help improve the accuracy and efficiency of breast cancer screening.

However, there is still a high uncertainty in the accuracy of mammography analysis by professional doctors, which is prone to misdiagnosis and missed diagnosis. The resulting diagnosis results will cause anxiety for patients and unnecessary invasive diagnostic procedures.

For the study, Google technical lead Shravya Shetty collaborated with researchers to train deep learning models for artificial intelligence using two datasets. One dataset contained 25,856 mammograms from the United Kingdom, and the other contained 3,097 mammograms from the United States. The results showed that the false positive rate of the detection results of the artificial intelligence model was 5.7% (US) and 1.2% (UK) lower than the typical radiologist, and the false negative rate was 9.4% (US) and 2.7% (UK) lower than the typical radiologist. .

The false positive rate, also known as the misdiagnosis rate, refers to the percentage of people who are actually not having a disease but are judged to have a disease based on a screening test. The false negative rate, also known as the missed diagnosis rate, refers to the percentage of people who actually have a disease but are determined to be disease-free based on a screening test.

New breakthrough! Google’s breast cancer AI detection system misdiagnosed only 5.7% lower than humans

Artificial Intelligence Breast Cancer Screening System

New breakthrough! Google’s breast cancer AI detection system misdiagnosed only 5.7% lower than humans

Performance of artificial intelligence systems and clinicians in breast cancer prediction According to data provided by the World Health Organization, breast cancer is one of the most common cancers in women, affecting approximately 2.1 million women each year. Worldwide, approximately 627,000 women died of breast cancer in 2018, accounting for approximately 15% of cancer deaths in women.

Breast cancer prediction, AI system compared with 6 independent radiographers In the independent sub-experiment of the above study, the AI ​​system outperformed all 6 radiologists.

Breast screening in the UK involves the analysis of mammograms by two radiologists. In response to this situation, the researchers found that using an artificial intelligence system could reduce the workload of a second doctor reading by 88 percent.

Breast cancer is not exclusive to women, men are also at risk of developing breast cancer and have a higher mortality rate.

In September 2019, a large-scale study published in JAMA Oncology, the top international oncology journal, showed that men with breast cancer had a 19% higher mortality rate than women. The study pointed out that male breast cancer patients were diagnosed at an older age than females, with an average diagnosis age of 63.3 years and an average female diagnosis of 59.9 years. Compared with female patients, male patients had higher mortality in all stages. The researchers concluded that clinical characteristics and undertreatment were associated with mortality in 63.3% of male patients.

Google Health, DeepMind, University College London, University of Cambridge, Royal Surrey Hospital in Guildford, Google-owned startup Verily Life Sciences, Stanford Medical Center, Royal Marsden Hospital, and more.

So far, there have been many attempts to diagnose breast cancer with artificial intelligence.

In 2017, Google Healthcare AI outperformed human professional pathologists in breast cancer diagnosis. The following year, Google released an AI detection system for advanced breast cancer that correctly distinguished metastatic cancer 99 percent of the time.

In October 2018, the Massachusetts Institute of Technology released a deep learning model for evaluating dense breast tissue in mammograms, and its diagnostic performance was similar to that of radiologists. In May 2019, MIT’s Computational Science and Artificial Intelligence Laboratory and Massachusetts General Hospital released an artificial intelligence system, claiming that it can predict whether a patient will develop breast cancer in the next five years from mammography images.

The Links:   LQ10D013 PM50B4LA060