Google “AI breast cancer detection accuracy is better than humans”

For breast cancer imaging, Google announced the findings that the AI system was able to detect cancerous tissue precursors with higher precision than human experts. In 2018, Google has developed an AI tool that detects metastatic breast cancer with a 99% probability and continues to research it.

The study confirmed that Alphabet’s DeepMind worked with Cancer Research UK Imperial Center, Northwestern University, and Royal Sally County Hospital to see if AI could support radiation and detect breast cancer signs more accurately. .

Breast cancer is a disease that can affect women around the world. In the UK, more than 55,000 people are diagnosed with breast cancer each year, and in the United States, it is reported that about 1 in 8 women develop it. If the disease is detected early and treated early, the possibility of cure is high, but on the other hand, it is difficult to accurately detect and diagnose.

Mammography, which uses a mammogram, is common as a breast cancer screening method. However, reading X-ray pictures is difficult even for experts. It can lead to false positives and false negatives, and problems such as delays in detection and treatment, unnecessary stress on the patient, and radiation workload.

The AI system announced this time is trained by anonymized mammograms among more than 76,000 women in the UK and more than 15,000 women in the United States, and targets non-identifying data sets of more than 25,000 in the UK and 3,000 in the US. It was an evaluation test.

Of course, the original learning data contained patient history, etc., for reference by human doctors. However, despite these limitations, the AI system’s false positives were 5.7% lower in the US and 1.2% lower in the UK than human experts. It is said that false negatives decreased by 9.4% in the US and 2.7% in the UK. A false positive test raises the risk of unnecessary treatment or surgery, while negative test mistakes are directly related to delay in treatment, so it can be said to be a sufficiently meaningful number.

The improved AI judgment accuracy and speed reduces patient waiting time and stress caused by misdiagnosis, and enables doctors to take appropriate action and treatment. However, continuous positive clinical studies and regulatory approval are required to improve precision and penetrate the medical field. Health information big data collection needs to be done carefully as there is also a personal information protection aspect. Related information can be found here .