Quantitative differentiation of benign and malignant mammographic circumscribed masses using intensity histograms

To investigate the usefulness of density gradient of mammographic masses for differentiating benign from malignant lesions, particularly circumscribed masses, which are difficult to diagnose by shape.

Phantom experiments were performed and diagnostic mammography examinations were reviewed. Mammograms of three acrylic resin globes differing in hardness were acquired with/without applied pressure, and density gradients were examined on intensity histograms with standard deviation (SD) as a hardness index. Similar analyses were performed using clinical mammographic examinations of circumscribed mass lesions. The usefulness of SD for differentiating between benign and malignant lesions was investigated by ROC curve analysis and minimum/maximum values of malignant and benign lesions, respectively.

For circumscribed masses (n = 196, benign, n = 176; malignant, n = 20), ROC analysis showed AUC = 0.786, with sensitivity = 70.0 %, specificity = 70.5 %, accuracy = 70.4 %, positive predictive value = 21.1 %, and negative predictive value = 95.4 % at SD = 64.46. Minimum and maximum SD of malignant and benign masses were 39.1 and 241.7, respectively.

On mammography, circumscribed masses can be diagnosed with moderate accuracy using the intensity histogram SD. Masses with SD below the minimum of breast cancer can be roughly diagnosed as clinically benign.