Simulation of mammographic lesions

Abstract

Rationale and objectives: This study presents a method for generating breast masses and microcalcifications in mammography via simulation. This simulation method allows for the creation of large image datasets with particular lesions, which may serve as a useful tool for perception studies measuring imaging system performance.

Materials and methods: The study first characterized the radiographic appearance of both masses and microcalcifications, examining the following five properties: contrast, edge gradient profile of masses, edge characteristics of masses, shapes of individual microcalcifications, and shapes of microcalcification distributions. The characterization results then guided the development of routines that created simulated masses and microcalcifications. The quality of the simulations was verified by experienced breast imaging radiologists who evaluated simulated and real lesions and rated whether a given lesion had a realistic appearance.

Results: The radiologists rated real and simulated lesions to have similarly realistic appearances. Using receiver operating characteristic analysis to characterize the degree of similarity, the results showed an A(z) of 0.68 +/- 0.07 for benign masses, 0.65 +/- 0.07 for malignant masses, and 0.62 +/- 0.07 for microcalcifications, thus showing notable overlap in the simulated and real lesion ratings.

Conclusion: This research introduced a new approach for simulating breast masses and microcalcifications that relied on anatomic characteristics measured from real lesions. Results from an observer performance experiment indicate that our simulation routine produced realistic simulations of masses and microcalcifications as judged by expert radiologists.