AI That Detects Breast Cancer Risk Has A 97 Percent Success Rate


Robin Andrews

Science & Policy Writer

The AI will likely be used to augment medical professionals, not replace them. Chompoo Suriyo/Shutterstock

It’s already been demonstrated that when it comes to pattern recognition, artificial intelligences (AIs) are far better at it than us mere mortals. Give them a game of Go, or challenge them to pick the most viable embryos during IVF treatment, and they’ve got us beat.

Now it appears that AI builders are turning their attention towards cancer – specifically, breast cancer, the most frequent type among women. A new conglomeration of code has been asked to determine whether a high-risk legion is malignant, and after being shown 335 examples, it has a 97 percent accuracy rate.


The AI was trained using the most cutting-edge information on breast cancer, including data on family histories, biopsies, pathology reports, recognizable lesion variations, and so on. With this knowledge made available to it, it appears that the program is more accurate than plenty of clinical physicians who rely more heavily on mammograms.

Over 1.5 million women are diagnosed with breast cancer every single year, and it’s responsible for 15 percent of all cancer deaths among this demographic. It’s worth pointing out, however, that mammograms tend to overreact to lesions that may not actually be cancerous, which has led to an overdiagnosis of breast cancer in some instances.

According to MIT’s Computer Science and Artificial Intelligence Laboratory, 90 percent of surgeries designed to remove high-risk lesions turn out to be benign after all. “This means that every year thousands of women go through painful, expensive, scar-inducing surgeries that weren’t even necessary,” they explain in a blog post.

Treating healthy patients for a non-existent cancer is not only detrimental to their bodies, but it’s a waste of resources that could be spent on someone who genuinely needs help in this regard.

Plenty of mammogram analyses and subsequent biopsies suggest a lesion could be cancerous when, post-surgery, it's revealed that this wasn't the case. Tyler Olson/Shutterstock

This AI – which is being shared by Harvard Medical School and Massachusetts General Hospital – is designed to make diagnoses more accurate and prevent these clinical errors from taking place. In fact, its incredible accuracy rate led to a reduction of 30 percent in the number of ultimately benign surgeries being performed.

It’s not about to replace human clinical practitioners though; rather, it’s designed to augment them. The understanding it obtained of breast cancer came from medical experts, after all, and any new information that needs to be added to its database can only be done by a human.

Unlike humans, however, an AI doesn’t get tired and make mistakes, nor does it suffer from stress. In this regard, it’s better at recognizing patterns than most practitioners using biopsies and deciding what to do with the data they’re presented with.

The results of this remarkable experiment were published in the journal Radiology.


[H/T: BBC News]


  • tag
  • breast cancer,

  • artificial intelligence,

  • prediction,

  • AI,

  • malignant,

  • accuracy,

  • mammogram,

  • lesions,

  • high-risk