By Kayla Matthews, HealthIT writer and technology enthusiast, Tech Blog
Twitter:Â @ProductiBytes
Discovering that you have a breast lesion is a scary thing. You wonder whether it could be cancerous and if you’ll need surgery.
Typically, doctors recommend an operation, even if the lesion might not be malignant. If it’s not, though, the unnecessary surgery could present more risk than benefit.
Through a partnership with MIT’s Computer Science and Artificial Intelligence Laboratory, researchers at Massachusetts General Hospital are working with a machine learning tool that could help prevent these unnecessary surgeries.
The technology analyzes how likely it is that a high-risk breast lesion will become cancerous, providing doctors with more clarity and patients with more peace of mind.
Cutting Down On Non-Critical Operations for Breast Cancer Patients
For this study, scientists focused on incidences where patients had already learned they have such lesions after undergoing needle biopsies. They found that the machine learning model successfully reduced unnecessary surgeries by one third, allowing more patients to follow intervention plans that entailed keeping a close eye on the identified issues rather than hastily scheduling the affected patients for operations.
Doctors typically recommend surgeries for people with this condition. Notably, that’s the usual course of action even when the breast problems do not pose immediate threats.
Thanks to this machine learning advancement, patients could experience fewer traumatic experiences stemming from non-essential operations.
Changing the Ways Doctors Treat Patients
Current methods of determining whether a growth is cancerous require doctors to make judgments once patients are in the middle of their surgical procedures. Plus, statistics say that 90 percent of the time, high-risk breast lesions are benign.
This means that patients have to go through costly and nerve-wracking surgeries, but machine learning may offer a better diagnostic method.
How Does Machine Learning Help?
The specific machine learning model used is called a random forest classifier. Researchers provided it with data from hundreds of cases involving the breast issue mentioned earlier. The information included details about the treatments those patients had received previously, their familial histories, demographic information and more.
To train the machine learning algorithms, scientists relied on patients who went through biopsies that detected high-risk lesions and had either had surgeries or two years worth of imaging after the initial diagnosis.
Overall, the machine learning technology indicated there were 1,006 high-risk lesions associated with patients in the study. However, only 115 of them (11 percent), became cancerous later. The researchers also discovered that certain factors about a patient’s case increased the likeliness of a cancer diagnosis. For example, when an individual’s pathology report was atypical, they were more likely to develop cancer in the future.
The model identified cancerous breast tumors 97 percent of the time, compared to the 79 percent accuracy rate associated with existing techniques. Experts say that because diagnostic tools are not exact, doctors tend to err on the side of caution and recommend surgical treatment. However, the results offered by machine learning could remove some of that uncertainty.
Instead of quickly recommending that patients have operations as soon as possible, physicians might take a different approach such as scheduling regular imaging appointments and making decisions based on what those pictures show.
Using Machine Learning to Determine Treatments
We’ve just looked at how scientists depend on machine learning to determine if a breast tumor requires an immediate operation. But, what can this technology do after people get the news that they have cancer?
A different study carried out by a research team at Western University showed positive outcomes when using machine learning to figure out treatments that will most likely be maximally effective for particular patients. It’s perpetually perplexing when two individuals with an identical kind of breast cancer respond drastically differently to the same treatment. Machine learning could lead to more personalized approaches to patient care.
It’s impossible to remove all the traumatic effects of receiving a diagnosis of breast cancer. However, as is clear from the research mentioned here, machine learning may significantly cut down on the number of times people unnecessarily get surgeries while making periodic imaging a more feasible solution.