Software for determination of a breast cancer

2019: Announcement

At the end of October, 2019 in hospitals began to use AI which defines a breast cancer with an accuracy of 90%.

Radiologists and programmers from the Center of data sciences of New York University developed new AI approach for confirmation of the diagnosis of a breast cancer which will allow to reduce the frequency of carrying out biopsies. Researchers found out that if to integrate possibilities of AI trained at more than one million x-rays with the analyses made by radiologists, the accuracy of diagnostics reaches 90%. If the trust of doctors to mammographic estimates increases up to such level, then need for frequent carrying out a biopsy will disappear by itself.

Hospitals began to use AI which defines a breast cancer with an accuracy of 90%.

Researchers reported that estimates of radiologists were exact in 78%, neuronet estimates - in 87%. The hybrid model of estimates of the radiologist on the basis of data of a neuronet provided the accuracy of diagnostics of 89%. At consolidation of several radiologists with a neuronet results were correct in 91% of cases.

Thus, new approach integrates capability of AI to detect changes of tissue of mammary gland at the level of pixels, invisible to a human eye, with forms of reasonings, unavailable AI. Such approach should reduce number of false positive and false-negative results that will increase trust of doctors to results of mammography.

The program was trained based on the data from 229,426 digital screening mammograms and 1,001,093 images collected within seven years of NYU Langone Health and analyzed by researchers. Images were compared with results of a biopsy, and estimated work of a new neuronet by the number of the correct forecasts. The command is going to increase diagnostics accuracy, having provided to the program more data — so a neuronet, perhaps, will even be able to reveal precancer statuses. ^[1]

Notes