50 Quick summary: A novel diagnostic platform combining lens-free holography and deep learning has automated breast cancer testing with 94.8% binary accuracy while costing less than US$980. The device utilizes an advanced neural network framework that flags low confidence predictions for expert review, significantly lowering diagnostic errors compared to traditional, expensive imaging scanners. For global healthcare practice and public policy, this affordable technology provides a scalable pathway towards expanding reliable cancer biomarker testing in resource limited areas, dramatically improving patient well-being through faster, equitable access to lifesaving treatments. Researchers at University of California, Los Angeles have developed a compact, low-cost diagnostic platform that combines lens-free holography with deep learning to automate HER2 scoring in breast cancer tissue samples. Their findings, published in BME Frontiers, demonstrate a potential alternative to conventional digital pathology systems, which are often expensive and difficult to deploy in resource-limited settings. Accurate assessment of HER2 status is a cornerstone of breast cancer diagnosis, prognosis, and treatment planning, particularly when determining eligibility for HER2-targeted therapies. Conventional whole-slide imaging scanners rely on complex optical components and precision mechanical systems, making them costly for many decentralised clinics and laboratories. The newly developed lens-free holography device operates without objective lenses or mechanical focusing. Using RGB laser illumination, it captures holographic diffraction patterns from stained tissue sections across a 1,250 mm² field of view, achieving an imaging throughput of 84 mm² per minute. According to the researchers, this throughput exceeds that of many commercially available pathology scanners, while the complete imaging hardware costs less than US$980. The system also achieved classification performance comparable to that of high-end brightfield microscopes used in standard digital pathology workflows. The team adopted an ensemble of five neural network models together with Bayesian Monte Carlo dropout for real-time uncertainty quantification. When evaluated on a blinded dataset of 412 independent tissue samples, the system achieved 84.% accuracy for 4-level HER2 classification and 94.8% accuracy for binary HER2 scoring. The uncertainty framework successfully identified 30.4% of misclassified samples while excluding only 7.2% of correctly classified predictions, helping reduce the risk of diagnostic errors. A simplified imaging mode using only blue laser illumination also maintained good classification performance while further lowering hardware costs. The neural networks reconstruct and analyse tissue features directly from raw holographic data, while the uncertainty quantification component flags low-confidence predictions for review by a human pathologist. HER2 status plays a central role in guiding treatment decisions, including the use of targeted therapies such as Trastuzumab, which can substantially improve outcomes for patients with HER2-positive breast cancer. However, access to reliable HER2 testing remains inconsistent in many resource-limited settings, contributing to disparities in diagnosis and treatment. The researchers suggest that the UCLA platform could help expand access to standardised HER2 assessment without requiring the substantial capital investment associated with conventional digital pathology infrastructure. The study was led by Aydogan Ozcan and colleagues Che Yung Shen, Xilin Yang, Yuzhu Li, Leon Lenk, and Zixiang Ji at University of California, Los Angeles. The platform was evaluated using immunohistochemically stained breast tissue sections, with its performance benchmarked against assessments by expert pathologists. The researchers note that further validation across diverse clinical settings will be necessary before the platform can be adopted widely. Integration with existing laboratory workflows, regulatory approval, and real-world clinical evaluation will also be important steps toward implementation. Beyond HER2 testing, the framework could potentially be adapted to assess other cancer biomarkers in digital pathology. If successfully validated and implemented, the technology could help expand access to faster and more affordable biomarker testing, particularly in resource-limited settings, supporting more timely and informed treatment decisions for patients with breast cancer.