Abstract
A BI-RADS category of 4 from a mammogram indicates suspicious breast lesions, which require core biopsies for diagnosis and have an approximately one third chance of being malignant. Human plasma contains many circulating microRNAs, and variations in their circulating levels have been associated with pathologies, including cancer. Here, we present a novel methodology to identify malignant breast lesions in women with BI-RADS 4 mammography. First, we used the miRNome array and qRT-PCR to define circulating microRNAs that were differentially represented in blood samples from women with breast tumor (BI-RADS 5 or 6) in comparison to controls (BI-RADS 1 or 2). Next, we used qRT-PCR to quantify the level of this circulating microRNAs in patients with mammograms presenting with BI-RADS category 4. Finally, we developed a machine learning method (Artificial Neural Network - ANN) that receives circulating microRNA levels and automatically classifies BI-RADS 4 breast lesions as malignant or benign. We identified a minimum set of three circulating miRNAs (miR-15a, miR-101 and miR-144) with altered levels in patients with breast cancer. These three miRNAs were quantified in plasma from 60 patients presenting biopsy-proven BI-RADS 4 lesions. Finally, we constructed a very efficient ANN that could correctly classify BI-RADS 4 lesions as malignant or benign with approximately 92.5% accuracy, 95% specificity and 88% sensibility. We believe that our strategy of using circulating microRNA and a machine learning method to classify BI-RADS 4 breast lesions is a non-invasive, non-stressful and valuable complementary approach to core biopsy in women with BI-RADS 4 lesions.