Fibroglandular Tissue Quantification in Mammography by Optimized Fuzzy C-Means with Variable Compactness

Abstract Background Mammography is a wordwild image modality used to diagnose breast cancer, even for asymptomatic women. Due to its large availability, mammograms can be used to measure breast density and to predict cancer development. Methods We developed a methodology to estimate breast density using post-processed digital mammogram. Our automatic approach utilizes an optimized Fuzzy C-Means with variable compactness algorithm to classify and quantify fibroglandular tissue in mammograms. Results Fibroglandular tissue percentage estimation by our method has been compared with BI-RADS assessment from radiologist and achieved 67.8% of correct classification, with Spearman's correlation coefficient of ρ = 0.618 , for p 0.001 . Furthermore, a Bland–Altman statistics showed no significant differences (bias of − 0.20 ± 1.52 ) between both methods, indicating that the assessment widely used in clinical routine is consistent with the results generated by the algorithms. Cohen's kappa coefficient comparing the performance of the algorithm with the visual assessment for the different BI-RADS scores was 0.47 suggesting a moderate agreement. Conclusion Then, our methodology showed to be robust and accurate when compared with visual assessment. Furthermore, our methodology is fully automatic and reproducible, avoiding inter and intra observers variation, which has a potential to be implemented in clinical routine.