Deep Learning Triage Cuts Breast Biopsy Turnaround by 38% in Simulation
A simulation study modelled the impact of a deep learning triage system on breast biopsy workflows in a Malaysian hospital setting. The CNN classifier achieved an AUC of 0.98, and discrete-event simulation projected a 38% reduction in turnaround time for suspicious cases, 22.5% less pathologist workload, and 15% savings on reagents and slides. Real-world validation is still required.
The original study
A Deep Learning Framework for Automated Triage of Breast Cancer Biopsies in Malaysia: A Simulation Study to Reduce Resource Consumption and Diagnostic Turnaround Time.
- Authors
- Kurniawan Budi Susilo Y, Yuliana D, Abdul Rahman S, Leong SL
- Journal
- Clinical breast cancer
- PMID
- 41863188
Original abstract
BACKGROUND: Breast cancer is a major health burden in Malaysia, where a shortage of pathologists causes long diagnostic delays, patient anxiety, and late treatment. Conventional histopathology workflows use a first-in-first-out (FIFO) system, which is inefficient since most biopsies are benign. This simulation-based study developed and validated a deep learning triage system to prioritize suspicious breast biopsy cases for pathologist review. MATERIALS AND METHODS: A convolutional neural network was trained on a large, ethically sourced synthetic dataset of whole-slide images, labelled as benign or suspicious (including atypical, in situ, and invasive carcinoma), and validated on an independent synthetic test set. A discrete-event simulation replicated the pathology workflow of a typical Malaysian hospital, comparing the deep learning triage system with standard FIFO reporting. Outcomes assessed included diagnostic turnaround time (TAT), pathologist workload, and laboratory resource use. RESULTS: The model achieved an area under the receiver operating characteristic curve of 0.98. Simulation showed a 38.2% reduction in average TAT for suspicious cases (7.2 to 4.5 days), with a small increase for benign cases. Pathologist workload fell by 22.5%, equivalent to saving 422 hours annually, while reagent and slide use declined by 15%. CONCLUSION: These in-silico findings project potential efficiency gains, though real-world validation is required. By expediting critical case reporting, reducing workload, and conserving resources, this approach offers a promising simulation-informed framework to address diagnostic bottlenecks in resource-constrained healthcare systems.