Machine learning helps Mass General predict cancerous breast lesions
Model identified 97 percent of lesions that were malignant and reduced unnecessary benign surgery by 30 percent.
Researchers at Massachusetts General Hospital, working with MIT’s Computer Science and Artificial Intelligence Laboratory, have developed a machine learning tool to identify high-risk breast lesions that are likely to develop into cancer.
By accurately predicting which biopsy-diagnosed, high-risk lesions are likely to become cancerous, the technology has the potential to reduce unnecessary surgeries by nearly one-third in this specific patient population. That’s the finding of a new study published online in the journal Radiology.
“This study is our proof of concept that we can actually change the way we’re managing our patients through machine learning algorithms,” says Constance Lehman, MD, senior author of the study and director of breast imaging at Massachusetts General Hospital.
Also See: Machine learning automatically identifies brain tumors
Researchers developed a machine learning model that analyzes traditional risk factors, such as patient age and lesion histology, as well as words that appear in the text of biopsy pathology reports. According to Lehman, the model was highly accurate when applied to a patient population with biopsy-diagnosed high-risk lesions who underwent surgery or at least two years of imaging follow-up.
“We identified 97 percent of the lesions that were malignant, and we reduced unnecessary benign surgery by 30 percent,” she notes. “So, we reduced our false positives and maintained an extremely high sensitivity.”
As a result of this successful proof of concept, Lehman says that she and her colleagues are working with the MGH Clinical Data Science Center to bring the machine learning model into the center’s clinical management algorithm.
“In the near future, we’ll be having the data run through our machine learning model so we can get the percentage of the risk that a particular lesion would or would not upgrade to cancer if we sent them to surgery,” she adds. “In this way, we think we can reduce the unnecessary benign surgeries and still maintain a very high rate of cancer capture.”
Lehman says radiologists want to be more targeted and precise in their recommendations to their patients. However, she contends that because high-risk breast lesions carry an increased risk of developing into cancer, surgical removal is often the preferred treatment option.
“We have not been nuanced, and we can certainly use more information available to us to make more informed decision-making with our patients” concludes Lehman. “We’re interested to see how machine learning can enhance the radiologist’s impact.”
“Because diagnostic tools are inexact, there is an understandable tendency for doctors to over-screen for breast cancer,” adds Regina Barzilay, co-author of the study and the Delta Electronics Professor of Electrical Engineering and Computer Science at MIT. “When there’s this much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent overtreatment.”
By accurately predicting which biopsy-diagnosed, high-risk lesions are likely to become cancerous, the technology has the potential to reduce unnecessary surgeries by nearly one-third in this specific patient population. That’s the finding of a new study published online in the journal Radiology.
“This study is our proof of concept that we can actually change the way we’re managing our patients through machine learning algorithms,” says Constance Lehman, MD, senior author of the study and director of breast imaging at Massachusetts General Hospital.
Also See: Machine learning automatically identifies brain tumorsResearchers developed a machine learning model that analyzes traditional risk factors, such as patient age and lesion histology, as well as words that appear in the text of biopsy pathology reports. According to Lehman, the model was highly accurate when applied to a patient population with biopsy-diagnosed high-risk lesions who underwent surgery or at least two years of imaging follow-up.
“We identified 97 percent of the lesions that were malignant, and we reduced unnecessary benign surgery by 30 percent,” she notes. “So, we reduced our false positives and maintained an extremely high sensitivity.”
As a result of this successful proof of concept, Lehman says that she and her colleagues are working with the MGH Clinical Data Science Center to bring the machine learning model into the center’s clinical management algorithm.
“In the near future, we’ll be having the data run through our machine learning model so we can get the percentage of the risk that a particular lesion would or would not upgrade to cancer if we sent them to surgery,” she adds. “In this way, we think we can reduce the unnecessary benign surgeries and still maintain a very high rate of cancer capture.”
Lehman says radiologists want to be more targeted and precise in their recommendations to their patients. However, she contends that because high-risk breast lesions carry an increased risk of developing into cancer, surgical removal is often the preferred treatment option.
“We have not been nuanced, and we can certainly use more information available to us to make more informed decision-making with our patients” concludes Lehman. “We’re interested to see how machine learning can enhance the radiologist’s impact.”
“Because diagnostic tools are inexact, there is an understandable tendency for doctors to over-screen for breast cancer,” adds Regina Barzilay, co-author of the study and the Delta Electronics Professor of Electrical Engineering and Computer Science at MIT. “When there’s this much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent overtreatment.”
More for you
Loading data for hdm_tax_topic #better-outcomes...