A computer program has shown a better accuracy of expert radiologists to identify cancers of the breast from images of mammography, according to a british study.
January 2, 2020 14h11
Breast cancer diagnosis: artificial intelligence can do better than the human
PARIS — A computer program has shown a better accuracy of expert radiologists to identify cancers of the breast from images of mammography, according to a british study.
Breast cancer is one of the most common cancers in women, with more than two million new cases diagnosed last year in the world.
These results have been published in the scientific journal Nature “suggests that we are in the process of developing a tool that can help doctors detect breast cancer with greater precision,” notes Dr Dominic King, head of uk at Google Health, and co-author of this study.
“Other tests, clinical validation and regulatory approvals are necessary before this can begin to make a difference for patients, but we are determined to work with our partners to achieve this objective”, adds the researcher, in a press release from the Imperial College of London.
This technique artificial intelligence (AI) resulting from the Google search is based on a mathematical model, an algorithm. The latter has been led, fed, with nearly 29, 000 images of mammograms from Great Britain and to a lesser extent in the United States.
The experts had access to patient history in the interpretation of radiographic images, while the AI had access to the last mammogram.
The AI has shown a reduction in the proportion of cases where a cancer has been detected wrongly, of 5.7 % on the images of american studied and 1.2 % on the british.
The algorithm has also reduced the percentage of diagnoses missed by 9.4 % between the images in the u.s. and 2.7 % among those from Great Britain.
“The more one identifies early breast cancer, the better it is for the patient”, said to AFP Dominic King, head of british of Google Health.
In the United States, a single reading of images of screening is usually performed, while in the United Kingdom, the routine mammography offered to women between 50 and 71 years were examined by two radiologists. It is also the case in the framework of the organized screening proposed in France for women aged 50 to 74 years.
The team of Google Health has also conducted experiments comparing the decision to the computer with the radiologist’s first reader.
If the two diagnoses agreed, the case was marked as solved. It is only in the event of discordant results than the one asked for then the camera to compare with the decision of the second player.
The study of King and his colleagues shows that the use of AI to verify the diagnosis of the first human reader could save up to 88% of the workload of the second radiologist.
“This technology represents an opportunity to support the excellent work done today examiners,” said Mr King.
The team hopes that this technology may one day serve as a “second opinion” for cancer diagnostics.