Abstract PD11-05: Intelligent shear-wave elastography to reduce unnecessary biopsies in breast cancer diagnosis (INSPiRED 002): An international, multicenter analysis

Abstract

Abstract Background: Breast ultrasound identifies additional carcinomas not detected in mammography, but has a higher rate of false-positive findings which result in more unnecessary breast biopsies. Shear-Wave Elastography (SWE), an ultrasound technique used to quantify the stiffness of a lesion, showed promising results to improve the diagnostic performance of B-mode breast ultrasound but also to miss some cancers. As the stiffness of a lesion is found to be influenced by individual patient characteristics, incorporation of lesion stiffness in more individualized assessments may be key to the problem of reducing unnecessary breast biopsies without impairing the breast cancer detection rate. Thus, in this study, we evaluated whether an intelligent algorithm incorporating traditional SWE values as well as other patient and clinical variables (hereafter “intelligent SWE”) could reduce the number of unnecessary breast biopsies without impairing the breast cancer detection rate compared to traditional SWE and B-mode breast ultrasound for patients with suspicious breast lesions. Methods: We trained, tested, and validated machine learning algorithms using patient, clinical, ultrasound, and SWE information to classify breast masses. We used international, multicenter data from 857 women with BI-RADS 4 breast masses at 12 study sites in 7 countries. Patients underwent B-mode breast ultrasound, SWE, and subsequent histopathologic evaluation. 10-fold cross-validation was used to train and test the algorithms on data from 11 of the 12 sites which were further validated using the additional site’s data. The results of B-mode breast ultrasound, traditional SWE, and intelligent SWE were compared to the gold standard of histopathologic evaluation. We calculated sensitivity, specificity, and AUROC and used McNemar tests to test for significant differences in diagnostic performance. Results: The mean age was 49.5 years (SD 16.3) and 42.2% breast masses (n=362 of 857) were found to be malignant as confirmed by histopathology. In the external validation set (n=285), traditional SWE showed a significantly higher diagnostic performance compared to B-mode breast ultrasound (P < 0.001), whereas intelligent SWE outperformed both B-mode breast ultrasound and traditional SWE (P < 0.001). The neural network algorithm showed a significantly higher diagnostic performance compared to the Logistic Regression with Elastic Net Penalty (P = 0.004). The neural network algorithm achieved a sensitivity of 100% (95% CI 97.1 to 100%, 126 of 126) and a specificity of 50.3% (95% CI 42.3 to 58.3%, 80 of 159); the number of unnecessary biopsies were reduced by 50.3% (79 vs. 159) without missing any cancer compared to B-mode breast ultrasound. Model-agnostic variable importance plots to provide insights into the model predictions showed that the three most important variables for intelligent SWE were patient age followed by Shear-Wave velocity and orientation of the lesion (parallel vs. not parallel) in B-mode ultrasound. Conclusion: This is the first evidence which suggests that the majority of false-positive breast biopsies could be safely avoided by using intelligent SWE without impairing breast cancer detection rates. These results may be helpful in their ability to reduce treatment burden for patients, providers, and healthcare systems. Trial registration: NCT02638935. Funding: Siemens Medical Solutions USA, Inc Diagnostic Performance ComparisonB-mode Breast UltrasoundTraditional Shear-Wave ElastographyIntelligent Shear-Wave Elastography – Logistic Regression with Elastic Net PenaltyIntelligent Shear-Wave Elastography – neural networkAUROC – value (95% CI)–0.84 (0.79-0.89)0.93 (0.90-0.95)0.93 (0.90-0.96)Sensitivity – % (95% CI); no.100% (97.1-100%); 126 of 12697.6% (93.2-99.5%); 123 of 126100% (97.1-100%); 126 of 126100% (97.1-100%); 126 of 126Specificity – % (95% CI); no.0% (0.0-2.3%); 0 of 15923.9% (17.5-31.3%); 38 of 15936.5% (29.0-44.5%); 58 of 15950.3% (42.3-58.3%); 80 of 159Negative predictive value – % (95% CI); no.–92.7% (80.1-98.5%); 38 of 41100% (93.8-100%); 58 of 58100% (95.5-100%); 80 of 80Positive predictive value – % (95% CI); no.44.2% (38.4-50.2); 126 of 28550.4% (44.0-56.8%); 123 of 24455.5% (48.8-62.1%); 126 of 22761.5% (54.4-68.2%); 126 of 205 Citation Format: André Pfob, Chris Sidey-Gibbons, Richard G. Barr, Volker Duda, Zaher Alwafai, Corinne Balleyguier, Dirk-André Clevert, Sarah Fastner, Christina Gomez, Manuela Goncalo, Ines Gruber, Markus Hahn, André Hennigs, Chi Ho, Panagiotis Kapetas, Sheng-Chieh Lu, Juliane Nees, Ralf Ohlinger, Fabian Riedel, Matthieu Rutten, Benedikt Schaefgen, Anne Stieber, Riku Togawa, Mitsuhiro Tozaki, Sebastian Wojcinski, Cai Xu, Geraldine Rauch, Joerg Heil, Michael Golatta. Intelligent shear-wave elastography to reduce unnecessary biopsies in breast cancer diagnosis (INSPiRED 002): An international, multicenter analysis [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr PD11-05.