Artificial intelligence (AI)-assisted digital pathology may be associated with a significant reduction in breast biopsy review time compared with standard digital pathology while maintaining high diagnostic accuracy.
In the retrospective, single-center, paired-reader CASCADE study, researchers evaluated 194 breast biopsy cases collected between January and July 2024 from adult female participants.At a pathology laboratory in Australia, two board-certified pathologists, each with more than 10 years of clinical experience, reviewed digitized hematoxylin and eosin–stained slides using standard digital pathology (DP) and AI-assisted digital pathology (DP+AI). Each case was reviewed under both conditions, with randomized review order and a 2-week washout period to reduce recall bias.
The researchers assessed pathologist review time, laboratory turnaround time (TAT), usability, diagnostic accuracy with and without AI assistance, and the standalone AI algorithm's performance in detecting breast cancer, invasive carcinoma, and ductal carcinoma in situ (DCIS). Diagnostic performance was assessed against original pathology diagnoses, with independent adjudication of identified discrepancies.
Mean pathologist review time decreased from 71.12 seconds with standard DP to 40.18 seconds with DP+AI. Both pathologists demonstrated shorter review times, with reductions of 36% and 52%, respectively. Review time decreased by 48% for DCIS and 47% for invasive carcinoma.
Cancer diagnostic agreement was 97% (n = 189/194) with standard DP vs. 99% (n = 193/194) with DP+AI. Cancer detection sensitivity was 98% without AI assistance and 100% with AI assistance, while specificity was 95% and 98%, respectively.
In the DCIS subgroup, diagnostic agreement increased numerically from 93% to 99%, while sensitivity increased from 86% to 100%. However, the analysis included just 21 DCIS-positive cases, limiting precision. For invasive carcinoma, sensitivity remained at 100% under both conditions, and diagnostic agreement was about 99% in each workflow.
The standalone AI algorithm demonstrated high diagnostic discrimination. Both participating pathologists rated the application positively, reporting that it supported diagnostic decision-making and increased confidence.
A separate retrospective analysis of 3,298 breast biopsy cases examined baseline laboratory TAT. Cases requiring immunohistochemistry (IHC) had a mean TAT of 3.73 days vs. 1.17 days for cases without IHC. Using these historical data, the researchers estimated that AI-guided IHC preordering may reduce TAT by about 2.5 days for applicable true-positive cases. The reduction was modeled rather than demonstrated through prospective implementation.
The study had several limitations, including its single-center design, involvement of just 2 pathologists, and assessment of 1 representative slide per case. Cases with inadequate image quality were excluded, and independent adjudication was limited to identified diagnostic discrepancies. The study was not powered to establish differences in diagnostic accuracy, and the small DCIS subgroup limited the reliability of subgroup comparisons. The proposed reduction in laboratory TAT remains hypothetical and requires prospective evaluation.
“These findings highlight the potential for integrating AI-enabled decision-support tools into routine breast pathology workflows to enhance efficiency, shorten TAT, and maintain diagnostic performance, although confirmation in larger multicenter studies is warranted,” wrote lead study author Jeremy Parry, of Western Diagnostic Pathology in Australia, and colleagues.
Full disclosures of the study authors can be found in the study.
Source: Journal of Pathology Informatics
