The aim of this study: To implement a spatial fuzzy C-means algorithm for image segmentation and breast tissue density quantification and compare it with BI-RADS breast density classes determined by radiologists.
Material and methods: The analysis was based on 206 mammograms performed in 111 women with various breast abnormalities. Digitized mammographic films were independently double read by radiologists certified in breast diagnosis, followed by consensus with arbitration agreement (radiological ground truth). Reporting was done using the BI-RADS mammography lexicon. Using an algorithm based on a combination of spatial fuzzy C-means clustering and binary thresholding, percent mammographic density was computed in digitized mammograms. The BI-RADS breast density readings were compared with percent breast density measurements determined by computer algorithm.
Results: The algorithm was found to match the BI-RADS density classification in 90% of the cases, with an excellent agreement (kappa = 0.88) between the radiological ground truth versus the algorithm breast tissue density estimates.
Conclusions: Our study proposed an algorithm that can be applied both to digitized and digital mammograms, which proved to be effective in breast density estimates. The method can accurately determine the percentage density removing the human observer variability. The proposed method showed an excellent agreement with radiological ground truth.