Home » News » Random Forest AI Technology Improves Breast Cancer Detection Rates

Random Forest AI Technology Improves Breast Cancer Detection Rates

The future of breast cancer diagnoses could soon be improved by the use of AI. Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts General Hospital, and Harvard Medical School are currently honing technology that has shown a 97% accuracy rate, an 18% improvement rate from without the use of AI, for diagnosing malignancies in high-risk lesions.

When faced with high-risk lesions, doctors and patients most often choose to have surgery, but 90% of the time the lesions are benign. Dr. Regina Barzilay recently published her findings about how AI can more accurately diagnose breast cancer in high-risk lesions along with co-writers with Dr. Constance Lehman and Dr. Manisha Bahl of MGH, as well as MIT grad students Nicholas Locascio, Dr. Lili Yu and Adam Yedidia.

Yedidia explains how the AI technique used, called Random Forests, is composed of trees of data points that create the likelihood of malignancy. “A single decision tree is composed of nodes and branches to other nodes. Each node contains a simple question—for example, is this patient over the age of 35? Another node downstream of it might ask, does this patient have a history of smoking? It’s basically a flowchart. If you answer yes to one question, you get to another question, and so on and somewhere at the bottom of these trees, it’s going to give you a probability of cancer. There’s a way to build these trees in such a way that they fit the data points you are looking at pretty well.”

In turn, all the trees comprise the random forest because using only a single tree would compromise the result. Yedidia explains, “The problem with using a single tree when if you give it all the data you are working with it might overfit your data. A single tree might over-extrapolate and try and really twist itself to fit those data points, and as a result it might not actually generalize very well to future data, to future patients who come in. You’d better served by taking a more conservative view. What fits the more general trend is this collection of many simpler trees, each of which only sees part of the full set of data, rather than a single complicated tree that tries to fit every single data point.” Hence, the name random forest refers to the fact that there are many trees, each of which has a say in the ascertained probability of cancer. Yedidia sums up the process: “you let the trees vote on it.”

Yedidia is optimistic about the future of AI and curing breast cancer. “It seems likely to me that AI is potentially very helpful for solving breast cancer. A lot of breast cancer is looking at images for tiny details that maybe humans aren’t necessarily good enough at noticing. A human might have trouble taking all these tiny details into account,but a computer can do it without a problem. Could AI in the near future be able to make some very accurate diagnoses in areas where humans have trouble? I think certainly yes. I think it’s very likely that AI in the next 10 years will have a lot to offer in cancer detection.”

Leah D'Sa

    Leah D’Sa is a Junior studying Writing, Literature, and Publishing at Emerson College. She is currently a copyeditor for the school newspaper the Berkeley Beacon as well as Poetry Editor for the literary magazine the Emerson Review. She is looking to begin her career with health technology writing as she seeks to combine her lifelong love of writing and science.

    Similar posts

    Leave a Reply

    Your email address will not be published. Required fields are marked *

    Follow us!

    Send this to a friend