Purpose: To evaluate the possibility of reducing unnecessary biopsies in patients with BI-RADS 3 lesions by implementing Kaiser score (KS).
Method: In this retrospective, single-center study, we included 79 female patients with BI-RADS 3 lesions and risk factors who underwent biopsy following magnetic resonance imaging. Three readers (two experienced breast radiologists and a radiology resident) blinded evaluated the lesions using KS. Lesions with a KS ≤ 4 were considered benign. Results were compared with the histopathological findings (gold standard) assessing sensitivity and specificity along with 95 % confidence intervals (CI) for each reader. Inter-reader agreement was assessed using Fleiss' kappa with 95 % CIs. Moreover, Cohen's kappa was used to assess concordance between two readers at time.
Results: 79 female patients (mean age, 50.9 ± 12.2 (standard deviation)) were evaluated. The area under the receiver operating characteristic curve for the three readers was excellent: 0.99, 0.99, and 0.90), respectively. The sensitivity of the two breast radiologists and the resident was 0.92 (95 % CI: 0.62 - 0.99), 1.00 (95 % CI: 0.95 - 1.00) and 0.75 (95 % CI: 0.42 - 0.95), respectively, while the specificity was 1.00 (95 % CI: 0.95---1.00), 0.99 (95 % CI: 0.92 - 1.00), and 1.00 (95 % CI: 0.95 - 1.00) respectively. By using a KS ≤ 4 value to indicate benignity, 66 to 67 biopsies (84 to 85 % of all the biopsies) would have been avoided. Inter-reader concordance via Fleiss' kappa was 0.792 (95 % CI: 0.68 - 0.91).
Conclusions: The implementation of KS could have spared 84-85% of biopsies, proving to be a quick, user-friendly tool with strong inter-observer agreement and high specificity.
Keywords: Biopsy; Breast neoplasms; Kaiser score; Machine learning; Magnetic Resonance Imaging.
Copyright © 2024 The Author(s). Published by Elsevier B.V. All rights reserved.