AI lensfree holography enables reliable automated HER2 assessment for breast cancer diagnostics
The Californer/10349153

LOS ANGELES - Californer -- The integration of AI into digital pathology has the potential to transform cancer diagnostics by enabling scalable and quantitative analysis of tissue specimens. However, widespread deployment of AI-assisted pathology remains challenged by the need for costly imaging infrastructure and the lack of reliable mechanisms to assess prediction confidence. Researchers at the University of California, Los Angeles (UCLA) have developed an uncertainty-aware computational pathology platform that combines lensfree holographic imaging with deep learning to perform automated HER2 assessment in breast cancer tissue samples.

HER2 (human epidermal growth factor receptor 2) is an important biomarker used in breast cancer diagnosis and treatment planning. Accurate HER2 scoring is essential because it directly influences therapeutic decisions and patient management. Conventional digital pathology workflows typically rely on high-end optical microscopy systems that require imaging lenses, mechanical focusing, and complex optical alignment. The UCLA-developed platform addresses these limitations through a compact lensfree holographic imaging system that captures diffraction patterns over a large field of view and reconstructs computational representations of tissue specimens for AI-based analysis.

More on The Californer

• Coastal Dental Arts Offers San Diego Patients Long-Lasting Dental Implants for a Confident Smile
• Southland Wind Ensemble 25-26 Season Finale
• 3C Gallery exhibit at The Reef Downtown LA
• What Would you Do with Your Time if it Was Actually Money?
• Plaza Mexico Presents 'Golazo Fest'

To improve diagnostic reliability, the researchers incorporated a confidence-aware deep learning framework together with Bayesian uncertainty quantification. The AI system not only predicts HER2 scores but also estimates the confidence of each prediction, enabling identification and flagging of uncertain cases for additional review. This uncertainty-aware approach provides an additional layer of safety for clinical decision support by reducing the impact of low-confidence predictions.

Using a blinded test set of 412 breast tissue samples, the proposed framework achieved 84.9% accuracy for four-class HER2 scoring and 94.8% accuracy for clinically relevant binary classification. The uncertainty-guided strategy further improved reliability by selectively identifying less certain predictions, resulting in an overall error correction rate of 30.4%. Despite its simplified optical hardware design, the lensfree imaging platform achieved performance comparable to conventional bright-field microscopy-based digital pathology systems while providing high-throughput imaging over a large sample area.

More on The Californer

• Mr. Hospital Bed Showcases the Best Hospital Bed and Air Mattress for Bed Sores for 2026
• The Packaging Company Expands Airless Pump Bottles for Cosmetic and Skincare Brands
• Traian TKD Tractari Auto Iasi: cum transporti legal la RAR o masina fara numere sau cu ITP expirat
• California and federal government secure $3 million in refunds for mortgage fraud victims
• Ajala Sen Selected as Long Beach's Fourth Youth Poet Laureate

"By combining computational imaging, deep learning, and uncertainty quantification, we aim to create accessible and trustworthy diagnostic technologies that can support pathology workflows in both advanced and resource-limited healthcare settings," said Prof. Aydogan Ozcan.

Beyond HER2 assessment, the proposed framework provides a scalable imaging–AI paradigm that could be extended to a wide range of biomarker evaluation tasks and digital pathology applications. By integrating uncertainty-aware AI with compact computational imaging hardware, this approach may help accelerate the adoption of trustworthy AI-assisted diagnostics for cancer detection, diagnosis, and treatment guidance.

The study was supervised by Prof. Aydogan Ozcan of UCLA. The other authors of this work include Che-Yung Shen, Xilin Yang, Yuzhu Li, Leon Lenk, and Zixiang Ji.

Article: https://spj.science.org/doi/10.34133/bmef.0278