News10NBC Investigates: Artificial intelligence a ‘game changer’ for breast cancer detection

[anvplayer video=”5177149″ station=”998131″]

ROCHESTER, N.Y. – Six months of chemo to shrink the tumor in her breast, then surgery to remove what was left of it.

Then six weeks of radiation, and more chemo, followed by a year-long clinical trial.

This was the cancer experience for News10NBC’s Deanna Dewberry.

But Dr. Kamal Kothari, a radiologist for 40 years, envisions a world where cancer is caught so early, treatment may require little more than a half hour outpatient surgery.

“It is a powerful technology. It is a game changer,” Dr. Kothari said.

She’s talking about the artificial intelligence being used at Borg & Ide Imaging that helps her find the tiniest tumor.

“It will highlight that area for us,” she said.

AI – when it detects cancer – highlights it on the mammogram.

Dr. Kothari shows us a recent mammogram at Borg & Idea. There are no highlighted areas, so AI has not detected cancer.

That is not the case with another mammogram she’s analyzing. The abbreviation ‘INT’ is next to a highlighted area, telling the radiologist there is an intermediate risk of cancer.

But on the other breast, “it’s suspicious,” Dr. Kothari said.

Highly suspicious, as indicated by the letters ‘HI’ by each box.

“Now this is a dense breast,” she explained.

Dense means that there’s fibrous and glandular tissue throughout the breast which shows up as white on mammograms. But cancer is also white, making it hard to find in a woman with dense breasts.

But not for AI.

Dr. Kothari: “This patient has a lump and that was cancerous,” she said.

Deanna Dewberry: “How helpful was AI for you as a doctor in reading this mammogram?”

Dr. Kothari: “In this one, I would say very helpful.”

That’s because the cancerous lump, which is white, is beneath fibrous healthy tissue, which is also white.

“That lump is hidden under the breast tissue,” Dr. Kothari said.

The AI software is the brainchild of Dr. Gregory Sorensen, a radiologist and former Harvard professor. He recently co-founded Deep Health. Through the new company, he created the AI technology that reads mammograms.

“It’s actually something that we humans struggle with. It’s finding the needle in the haystack,” he said.

His company essentially trained his algorithm to find that proverbial needle in a haystack.

“The way modern AI works is you show it a thousand dogs and a thousand cats and then it can learn to distinguish dogs from cats,” Dr. Sorensen said.

In the same way, he taught AI to distinguish cancer from non-cancer.

Statistically, we know a busy radiologist like Dr. Kothari will likely see 2,000 cancer cases over a 40-year career.

“So our A.I. has seen four lifetimes of cancers. We trained it on 8,000 cancers to start with,” Dr. Sorensen said.

And have added thousands more since his software was put to use at more than 350 imaging centers across the county. While AI doesn’t replace a skilled radiologist, it does give doctors another trained set of eyes.

Dr. Kothari: “These days we are diagnosing breast cancer when they are in situ… . That means they’re still inside the cell. They haven’t ruptured the membrane yet.”

Dewberry: “ Oh my gosh. Okay that gave me chills.”

And that very early detection could mean breast cancer is no longer something to be feared: rather, something that is nearly always curable.

“We have found literally hundreds of cancers that would have been missed without the AI. It’s so gratifying,” Dr. Sorensen said.

AI is far from perfect at this point.  Dr. Kothari had one example of a mammogram in which AI was correct in finding cancer in one spot and was wrong in another. That’s why AI does not replace radiologists, but it is a valuable tool.

Right now, in Rochester, AI is being used at Borg and Ide, Elizabeth Wende and URMC.  Rochester General Hospital is adding it soon.  Borg and Ide charges $60 for the additional AI reading because insurance doesn’t yet cover it.  Researchers in the field believe that in the near future, AI will be standard in mammography.