Radiologists leverage AI to characterize cancerous breast lesions

Artificial Intelligence (AI) can now determine whether high-risk breast lesions (HRLs) are likely to become cancerous, in turn avoiding unnecessary surgery in nearly one-third of patients, according to a new study by researchers from Massachusetts General Hospital and Harvard Medical School in Boston. 

“Up to 14 percent of image-guided biopsies performed on the basis of suspicious mammograms yield high-risk breast lesions,” the authors, led by Manisha Bahl, MD MPH, director of the Breast Imaging Fellowship Program, wrote. “Most HRLs are benign, but surgical excision typically is recommended because of the low but present potential for upgrade to ductal carcinoma in situ or invasive malignancy at surgical excision. The resulting status quo is overtreatment with unnecessary surgery for HRLs that are not associated with malignancy.” 

The study was published online Oct. 17 in Radiology. Researchers trained the machine learning model, the random forest classifier, on a group of patients with biopsy-proven HRLs who already had surgery or at least two years of imaging follow up. 

  • 1,006 HRLs were identified from the 986-patient study cohort. 
  • 303 (30.1) percent of core biopsies showed more than one HRL. 
  • The machine learning model was developed using 671 of the HRLs and then tested with the remaining 335 HRLs. 
  • 115 (11.4 percent) of the high-risk lesions were upgraded to cancer. 

Thirty-seven 37 of 38 malignancies (97.4 percent) were diagnosed at surgery, and 30.6 percent of surgeries of benign lesions could potentially have been avoided. 

“This study provides proof of concept that a machine learning model can be applied to predict the risk of upgrade of HRLs to cancer,” Bahl et al. wrote. “Use of this model could decrease unnecessary surgery by nearly one-third and could help guide clinical decision making with regard to surveillance versus surgical excision of HRLs.” 

Future work includes incorporation of mammographic images and histopathologic slides into the machine learning model.