Abstract PO2-28-07: Artificial Intelligence-based Computer-aided Diagnosis for Mammography: Predicting Occult Invasive Cancer in Women with Biopsy-proven Ductal Carcinoma in Situ

Abstract Background: Ductal carcinoma in situ (DCIS) is a noninvasive breast cancer confined within the basement membrane. Due to the increase in breast cancer screening, DCIS is increasingly being diagnosed on biopsy. Active surveillance is a novel and evolving treatment strategy for DCIS to reduce overtreatment. Currently ongoing clinical trials of active surveillance include only mammographic findings in their eligibility criteria, however mammography has limited ability to detect occult invasive cancer. The objective of this study was to investigate whether an artificial intelligence-based computer-aided diagnosis (AI-CAD) application for mammography can improve the prediction of occult invasive cancer in women with percutaneous needle biopsy-proven DCIS. Methods: A retrospective search of our database identified consecutive women with percutaneous needle biopsy-proven DCIS who underwent surgery at Seoul National University Hospital (Seoul, Korea) between June 2019 and May 2021. Two board-certified breast radiologists reviewed preoperative mammographic findings of the primary tumor in consensus according to the Breast Imaging Reporting and Data System. Using a commercially available AI-CAD (Lunit INSIGHT for Mammography, Lunit Inc.), quantitative abnormality scores of 0 to 100% of the ipsilateral breast were obtained. Data were collected on age at diagnosis, symptoms, biopsy methods, DCIS grade on biopsy, and presence or absence of invasive cancer and axillary lymph node metastases on surgical pathology. Multivariable logistic regression models were used to evaluate factors associated with prediction of occult invasive cancer. Results: Of 692 biopsy-proven DCIS in 692 women (mean age, 53 years ± 11 [standard deviation]), 298 (43%) presented as calcifications only, 160 (23%) as mass or asymmetry, and 111 (16%) as masses with calcifications. The other 123 (18%) cancers were not demonstrable on mammography. Surgical pathology revealed occult invasive cancer in 278 (40%) women and 18 (3%) had axillary metastasis. Women with occult invasive cancer showed higher abnormality scores on AI-CAD than women with pure DCIS (mean, 70% vs. 51%; P < .001), and women with axillary metastasis had higher abnormality scores on AI-CAD than women without axillary metastasis (mean, 88% vs. 58%; P = .002). High ( >75%) abnormality score on AI-CAD (Odds ratio [OR], 1.8 [95% CI: 1.3, 2.6]; P < .001), age of 55 years or higher (OR, 1.6 [95% CI: 1.2, 2.3]; P =.005), present symptoms in the ipsilateral breast (OR, 2.7 [95% CI: 1.9, 4.0]; P < .001), biopsy performed with a 14-gauge automated gun (OR, 2.5 [95% CI: 1.7, 3.6]; P < .001), and high-grade DCIS at biopsy (OR, 2.5 [95% CI: 1.7, 3.5]; P < .001] were independent predictors of occult invasive cancer with a C-index of 0.75 (95% CI: 0.72, 0.78). Conclusion: The quantitative score of preoperative mammography obtained from a commercially available AI-CAD was able to predict occult invasive cancer in an easy and simple manner. AI-CAD applications for preoperative mammography may be useful in predicting occult invasive cancer in women with biopsy-proven DCIS, especially in those without preoperative breast MRI. Citation Format: Su Hyun Lee, Hajung Kim, Su Min Ha, Soo-Yeon Kim, Jung Min Chang, Nariya Cho, Woo Kyung Moon. Artificial Intelligence-based Computer-aided Diagnosis for Mammography: Predicting Occult Invasive Cancer in Women with Biopsy-proven Ductal Carcinoma in Situ [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO2-28-07.