Researchers at the University of California, Los Angeles have developed a compact, cost-effective diagnostic platform combining lensfree holography and deep learning for automated HER2 scoring of breast cancer tissue samples. Their findings have been officially published in BME Frontiers (BMEF), presenting a transformative alternative to expensive and bulky conventional digital pathology equipment for clinical use.
Accurate HER2 evaluation is a cornerstone of breast cancer diagnosis, prognosis and targeted therapy planning. Traditional whole-slide imaging scanners depend on sophisticated optical components and precise mechanical systems, making them unaffordable for many decentralized clinics. The newly designed lensfree holography device requires no objective lenses or mechanical focusing. Under RGB laser lighting, it captures holographic diffraction signals from stained tissue sections over a field of view of 1,250 mm², with an effective imaging throughput of 84 mm² per minute, outperforming many commercial pathology scanners.
To ensure dependable results, the team adopted a five-model neural network ensemble strategy and Bayesian Monte Carlo dropout for real-time uncertainty quantification. Evaluated on a blinded dataset of 412 independent tissue samples, the system reached 84.9% accuracy for four-level HER2 classification and 94.8% accuracy for binary scoring. It successfully filtered out 30.4% of misclassified samples while only losing 7.2% of correct predictions, effectively lowering diagnostic risks. A simplified single blue-light mode also maintained decent performance and further reduced hardware expenses.
The complete imaging hardware costs less than $980, striking an excellent balance between affordability and functionality. Its overall classification performance is comparable to high-end brightfield microscopes used in standard digital pathology workflows.
This all-new imaging-AI framework offers a practical solution for high-throughput, on-site HER2 testing. It will expand access to standardized breast cancer pathology services, especially in regions short of advanced medical equipment, and drive the popularization of low-cost computational pathology technologies.
