A new machine learning model allows for physicians to determine whether atypical ductal hyperplasia (ADH) could upgrade to cancer, according to new research published in JCO Clinical Cancer Informatics. The model can identify 98 percent of all malignant cases prior to surgery, while sparing 16 percent of women from unnecessary surgeries on benign lesions.
ADH is a breast lesion that increases the risk of breast cancer by four- to five-fold. Typically, ADH is found with mammography and its presence confirmed using biopsy. Previous research published in Current Problems in Diagnostic Radiology found that 95 percent of breast imagers recommend surgical removal for all ADH cases discovered during biopsy to establish if the lesion is cancerous.
“The excision of all ADH lesions may lead to overtreatment, which results in invasive surgeries for benign lesions in many women,” wrote lead author Saeed Hassanpour, PhD, assistant professor of biomedical data science and epidemiology at Dartmouth University, and colleagues. “A machine learning method to predict ADH upgrade may help clinicians and patients decide whether combined active surveillance and hormonal therapy is a reasonable alternative to surgical excision.”
Hassanpour and colleagues sought to find a machine learning algorithm that could help physicians and patients determine whether active surveillance and hormonal therapy could replace surgery.
The researchers assessed 128 lesions from 124 women at an academic care center in New Hampshire who exhibited ADH in a biopsy who underwent surgery from 2011 to 2017. They developed six different machine learning models to calculate ADH upgrade from core needle biopsy—gradient-boosting trees, random forest, radial support vector machine (SVM), weighted K-nearest neighbors (KNN), logistic elastic net and logistic regression.
The best performing models were gradient-boosting trees with 78 percent accuracy and random forest with 77 percent accuracy. Furthermore, the top important characteristics that determined ADH upgrade to cancer were: age at biopsy, the size of the lesion, the number of biopsies, the needle gauge and personal/family history of breast cancer.
According to the researchers, the random forest model could have diagnosed 98 percent of malignancies via surgical biopsies and spared 16 percent of women from unnecessary surgeries on benign lesions if used.
"Our model can potentially help patients and clinicians choose an alternative management approach in low-risk cases,” Hassanpour said in a prepared statement. "In the era of personalized medicine, such models can be desirable for patients who value a shared decision-making approach with the ability to choose between surgical excision for certainty versus surveillance to avoid cost, stress and potential side effects in women at low risk for upgrade of ADH to cancer."
The team hopes to develop the scope of their machine learning model to include other high-risk breast lesions. Additionally, they plan on further validating their model on external datasets using state-level and national breast cancer registries.