Breast cancer is one of the most common cancer among American women. In an year around 40,000 women die from breast cancer in the U.S. alone. The symptoms of breast cancer include painless breast lumps, nipple discharge, and inflammation of the skin of the breast. There are few test like Mammograms are available to detect this cancer at early stage, but it is still imperfect and often shows false positive results.
Bejnordi and his colleagues from the Radboud University Medical Center in Nijmegen has the solution of this problem. They have created an AI system that can detect metastases in hematoxylin and eosin-stained tissue sections of lymph nodes of women with breast cancer. It’s deep learning algorithm has high sensitivity and specificity for identifying diabetic retinopathy according JAMA.
“AI is increasingly being recognized as a major element of the healthcare landscape. We are now at a turning point where AI algorithms perform as well as or better than clinicians at specific tasks. But still, I did not expect such remarkable results at this early stage. We showed that state-of-the-art AI algorithms perform as well as or better than pathologists in detecting the spread of breast cancer to lymph nodes,” Babak Ehteshami Bejnordi, an author of the study, told Healthline.
Thirty-two algorithms were developed during the international Cancer Metastases in Lymph Nodes (CAMELYON16). The challenge was to create algorithms that could detect the spread of breast tumor cells to nearby lymph nodes, which is important in estimating a woman’s prognosis.
“To build the system,” Bejnordi explained, “the deep learning algorithm is exposed to a large dataset of labeled images, and it teaches itself to identify relevant objects.”
In study, the researchers performance the deep learning algorithms against the performance of 11 pathologists. This algoritm independently analyzed 129 digitized images of patients’ lymph nodes. The doctors were given a time limit to accomplish the task. In a separate test, the algorithms were pitted against one pathologist who was free of time constraints.
It turned out that some algorithms bested the pathologists who were under time limits. The pathologist without time constraint detect images at a mean of 0.0125 false-positives per normal whole-slide image. The researchers found that the area under the receiver operating characteristic curve (AUC) ranged from 0.556 to 0.994 for the algorithms.
“We are now at a turning point where computers perform better than clinicians at specific tasks,” Bejnordi said.
Bejnordi acknowledged the study’s limitations, and said the technology has to be tested in real-world practice. But in general, he said, the health-care field is increasingly seeing the potential of artificial intelligence. Bejnordi says the use of AI in pathology could take a lot of pressure off specialists.
Both studies were published in the Journal of the American Medical Association.
Read also, you may like it :