Salesforce claims AI can detect signs of breast cancer with 92% accuracy

Salesforce has peeled off the curtains today ReceptorNet, Lawrence J. Ellison, University of Southern California, a machine learning system researcher at the USC Institute for Transformational Medicine, developed in collaboration with clinicians. The system, which can determine biomarkers important to oncologists in determining the appropriate treatment for breast cancer patients, achieved 92% accuracy in a study published in the journal. Nature Communications..

Breast cancer affects more than 2 million women each year, and one in eight women in the United States develops breast cancer for life. In 2018 in the US alone 2,550 new cases of breast cancer in men..And the incidence of breast cancer increase In almost every region of the world.

To address this, Salesforce researchers have developed an algorithm (ReceptorNet, mentioned above) that can predict the state of hormone receptors from images of inexpensive and ubiquitous tissues. Breast cancer cells extracted during biopsy or surgery are usually tested to see if they contain proteins that function as estrogen or progesterone receptors. (When the hormones estrogen and progesterone attach to these receptors, they promote the growth of cancer.) However, these types of biopsy images are not very widely available and should be reviewed by a pathologist. ..

In contrast to the immunohistochemical processes preferred by clinicians, who require a microscope and tend to be expensive and not readily available in parts of the world, ReceptorNet accepts hormones by shape-sensitive hematoxylin and eosin (H & E) staining. Determines the condition of the body. Size, and cell structure. Salesforce researchers have trained the system with thousands of H & E image slides from cancer patients in “dozens” hospitals around the world.

Above: A descriptive interpretation of how AI finds things invisible to the human eye.

Image Credit: Salesforce

Studies show that much of the data used to train algorithms for diagnosing illness can perpetuate inequality.Recently, a team of British scientists found Almost all eye disease datasets are from patients in North America, Europe, and China, and it is uncertain whether eye disease diagnostic algorithms will work well in racial groups in underestimated countries. In another study, researchers at Stanford University identified most of the US data on studies, including medical use of AI, from California, New York, and Massachusetts.

However, Salesforce says that when it analyzed ReceptorNet for signs of bias along age, race, and geographic vectors, it found that there was no static difference in performance. They also state that they provided accurate predictions regardless of the difference in the preparation of the tissue samples analyzed.

Salesforce believes that clinical implementation of systems like ReceptorNet can reduce treatment costs and the time it takes to start breast cancer treatment, improve accuracy, and improve patient health. In the short term, ReceptorNet will lay the foundation for future research comparing the clinical workflows of pathologists with and without this type of AI. This can help clarify that possibility better.

In addition to Salesforce, there are many major tech companies that have been criticized for investing in AI that allows oncologists to diagnose cancer as reliably as possible. In January, Google Health, a branch of Google focused on partnerships in health-related research, clinical tools, and healthcare services, AI model According to the company, they were trained on over 90,000 mammogram x-rays and performed better than human radiologists. Google claimed that the algorithm could recognize more false negatives (images that look normal but include breast cancer) than previous studies, but some clinicians, data scientists, and engineers said in a statement. I have a problem with. In a counter-argument, the co-authors stated that the lack of detailed methods and code in Google’s research “damages its scientific value.”