Better breast cancer diagnosis via computer

You know that the days of human travel agents, taxi drivers and switchboard operators are numbered, but you probably thought that that the jobs of highly-skilled pathologists weren't in danger from the coming AI apocalypse.

You might be wrong.

Scientists at Case Western Reserve University have created a deep-learning computer network that was 100 percent accurate in determining whether invasive forms of breast cancer were present in whole biopsy slides.

The network correctly made the same determination in each individual pixel of the slide 97 percent of the time, rendering near-exact delineations of the tumors. The machine was more consistent and accurate than the analyses of four pathologists, in many cases improving on their delineations.

Cancer is present in 10 percent of every biopsy ordered by a doctor. Every biopsy must be analyzed by a pathologist to identify the volume and extent of the disease, determine if it has metastasized and ascertain whether the patient has an aggressive cancer and needs chemotherapy or some less drastic treatment.

Anant Madabushi, co-author of the study, doesn't see AI replacing human pathologists... yet. "If the network can tell which patients have cancer and which do not, this technology can serve as triage for the pathologist, freeing their time to concentrate on the cancer patients," he said.

Every AI has to be trained, and to educate the neural network at CWRUth scientists downloaded 400 biopsy images from multiple hospitals. Each image was approximately 50,000 by 50,000 pixels.

Once their software was up to speed on what cancer looks like, the scientists then presented the network with 200 images from The Cancer Genome Atlas and University Hospitals Cleveland Medical Center. The network scored a perfect 100 percent on determining the presence or absence of cancer on whole slides and nearly as high per pixel. Network training took about two weeks, and identifying the presence and exact location of cancer in the 200 slides took about 20 to 25 minutes each.

It is worth noting that the training of the AI was done two years ago, and the process took two weeks. With the advancement in computer processing speed over the past 24 months, Madabhushi estimates that training now would take less than a day, and cancer identification and delineation could be done in less than a minute per slide.

"To put this in perspective," Madabhushi said, "the machine could do the analysis during 'off hours,' possibly running the analysis during the night and providing the results ready for review by the pathologist when she/he were to come into the office in the morning."

The research has been published in Scientific Reports.