AI-Assisted Mammography Screenings Show Preliminary Detection Improvement
The assistance of AI-based computer-aided detection for mammogram screenings did not impact recall rates in a radiologist’s standard single reading.
The assistance of AI-based computer-aided detection for mammogram screenings did not impact recall rates in a radiologist’s standard single reading.
Preliminary results from the AI-STREAM study (NCT05024591) showed that breast radiologists using artificial intelligence (AI)–based computer-aided detection (AI-CAD) significantly improved cancer detection rates (CDRs) in screening mammograms vs screening without AI-CAD, according to a study published in Nature Communications.1
Data revealed that breast radiologists using AI-CAD (n = 140) elicited a CDR of 5.70% (95% CI, 4.76%-6.65%), which was significantly higher by 13.8% compared with those not using AI-CAD (n = 123) at 5.01% (95% CI, 4.13%-5.89%; P <.001). Furthermore, no changes in recall rates (RRs) were observed in the AI-CAD group (n = 1113) vs the control group (n = 1100), which elicited RRs of 4.53% (95% CI, 4.27%-4.79%) and 4.48% (95% CI, 4.22%-4.74%), respectively (P = .564). Another metric, PPV1, defined as the percentage of all positive screening exams with a pathologic cancer diagnosis within 1 year, favored the AI-CAD group at a value of 12.6 vs 11.2 with control (P <.001).
Further results showed AI-CAD detected an additional 6 cases of ductal carcinoma in situ (DCIS) and 11 cases of invasive cancer, leading to significant increases in both DCIS (P = .009) and invasive cancer detection (P <.001). AI-CAD assistance also significantly increased the detection of small-sized cancers of less than 20 mm (P = .002), node-negative metastasis (P = .001), luminal A subtype (P = .002), and lower-grade invasive ductal carcinoma of no special type (IDC NOS; P =.009) vs screening without AI-CAD.
“[G]iven the diverse mammography interpretation procedures across countries worldwide, there is a need to demonstrate the true positive impact of increased CDRs when AI is applied in various ways in real-world clinical environments,” Yun-Woo Chang, MD, of the Department of Radiology at Soonchunhyang University Seoul Hospital in Seoul, South Korea, wrote in the publication with study coinvestigators.1 “The preliminary results from this prospective AI-STREAM study demonstrated positive potential that AI assistance in radiologists’ interpretation is indeed beneficial for both [breast radiologists] and [general radiologists] in a single-reading strategy. With the assistance of AI-CAD, [breast radiologists] improved CDR and increased early cancer detection without affecting RRs in a single-reading strategy.”
The prospective, population-based study enrolled patients 40 years and older from 6 academic hospitals participating in a South Korea–based national breast cancer screening program between February 1, 2021, and December 31, 2022. Those with a history of breast cancer or mammoplasty were excluded from enrollment, as the AI software was not validated for these subgroups.
During breast cancer screening, mammography was conducted by technologists, and a single-reading strategy was employed using 2 standard craniocaudal and mediolateral oblique views of individual breasts using digital mammography, a standard procedure for interpreting mammograms in South Korea.
Patients diagnosed with screen-detected breast cancer (n = 140) had a median age of 62 years (IQR, 51-69), with 35.7% of patients ages 60 to 69 and 22.9% being 70 and older. Most patients had a breast density of C (68.6%). The overall patient population of those included in the national screening program (n = 24,543) had a median age of 61 years (IQR, 51-68), and 51.2% had a breast density of C.
The coprimary end points of the study were CDRs and RRs of breast radiologists with and without AI-CAD in mammography reading. Secondary end points included comparisons of diagnostic accuracy, including differences between general radiologists with and without CAD, breast radiologist arbitration reading and breast radiologist with CAD, and breast radiologist without CAD vs stand-alone AI-based CAD, among others. PPV1 was also assessed.2
References
- Chang YW, Ryu JK, An JK, et al. Artificial intelligence for breast cancer screening in mammography (AI-STREAM): preliminary analysis of a prospective multicenter cohort study. Nat Commun. 2025;16:2248. doi:10.1038/s41467-025-57469-3
- Artificial Intelligence for breaST canceR scrEening in mAMmography (AI-STREAM). ClinicalTrials.gov. Updated September 28, 2023. Accessed March 6, 2025. https://tinyurl.com/mv472zc9
