Computer-aided preoperative diagnosis of microcalcifications on mammograms.

Purpose: To evaluate of a computer-aided method for differentiating malignant from benign clustered microcalcifications.

Material And Methods: Our material was 350 suspicious microcalcifications on mammograms from 330 female patients who underwent breast biopsy (after hook wire localization and under mammographic guidance). The histologic findings were malignant in 140 cases (40%) and benign in 210 cases (60%). Those clusters were manually detected, computer-aided analyzed and quantitatively estimated. Besides computer analysis, 3 physicians-observers (2 radiologists and 1 breast surgeon) evaluated the malignant or benign nature of the clustered microcalcifications. The performance of the artificial network, each observer and the three observers as a group was evaluated by receiver operating characteristics (ROC) curves.

Results: Comparison of the ROC curves revealed the following AUC values (area under the curve): computer - 0.950, physician 1 - 0.815, physician 2 - 0.830, physician 3 - 0.830, and physicians as a group - 0.825. The results, compared by the student t-test for paired data, showed a statistically significant difference between computer analysis and physicians' performance, independently and as a group.

Conclusion: Our study showed that computer analysis achieved statistically significantly better performance than that of physicians in the classification of malignant and benign calcifications.