"BlurryScope": A Compact, AI-Powered Microscope for Rapid Cancer Scoring
The Californer/10323480

LOS ANGELES - Californer -- A research team at UCLA, led by Professor Aydogan Ozcan, has introduced BlurryScope, a compact, cost-effective scanning microscope that combines simple optical hardware with advanced deep learning algorithms to assess HER2 status in breast cancer tissue samples.

Digital pathology has become an indispensable part of modern cancer diagnostics, enabling precise evaluation of biomarkers such as HER2, a protein that plays a critical role in guiding treatment decisions for breast cancer patients. However, the high cost and large footprint of whole-slide imaging scanners have limited their widespread adoption, especially in clinics with constrained budgets or in resource-limited regions. Conventional scanners can cost upwards of $100,000, and their large size makes them impractical for many laboratories. In contrast, the BlurryScope system can be built for less than $650, occupies only a 35 x 35 x 35 cm space, and weighs just 2.26 kg—making it portable, accessible, and affordable.

More on The Californer

• Car Wash Deal in Downey CA Prices Starting from $8
• Dr. Vincent Malfitano Elected to Central Council of The Italian Catholic Federation, National Gov
• Your Body Isn't Broken—It's Out of Balance: The New Book Revealing the Blueprint to Restore Hormone Balance, Sleep, Gut & Metabolic Health
• Long Beach Recognizes September as National Preparedness Month, Relaunches Program to Connect Neighborhoods to Disaster Planning Resources
• SEEAG Adds Santa Barbara County Office, Student & Community Agriculture Learning Programs

BlurryScope operates by continuously scanning tissue slides, resulting in motion-blurred images that would normally be discarded. A specially trained deep neural network then interprets these blurred images to classify HER2 expression, effectively converting optical "imperfections" into a valuable diagnostic resource. This approach drastically simplifies the microscope hardware requirements, reducing both the cost and mechanical complexity of the device, while still delivering clinically relevant information.

In blinded experiments involving 284 unique patient tissue cores, the system achieved nearly 80% accuracy across the standard four HER2 scoring categories (0, 1+, 2+, 3+). When the scores were grouped into two clinically actionable categories (0/1+ versus 2+/3+), accuracy increased to almost 90%. Importantly, repeated scans of the same samples demonstrated strong reproducibility, with more than 86% of classifications matching across multiple independent runs. These results highlight not only the robustness of the AI model but also the reliability of the entire system for real-world clinical use.

More on The Californer

• California claims victory – again – over Huntington Beach as appeals court rules against city's NIMBY violations of state law
• The Great Junk Hunt Returns to Roseville – A Vintage Shopping Adventure Awaits!
• Erdos Technologies CEO Dan Herbatschek Emphasizes the Importance of Software in Helping Companies Make More Intelligent Decisions on Inventory Flow and Management
• MyNamePlayer Empowers Educators to Create Inclusive Classrooms Through Correct Name Pronunciation
• Long Beach Seeks Volunteers for 41st Annual California Coastal Cleanup Day, September 20

Beyond its performance metrics, BlurryScope automates the entire diagnostic workflow—from continuous scanning and image stitching to region-of-interest cropping and classification—making it a turnkey solution for pathology laboratories. While the device is not intended to replace high-end digital pathology scanners, it provides a powerful complementary tool for rapid triage, preliminary assessments, or use in settings where traditional systems are impractical or unavailable.

The potential impact of BlurryScope extends well beyond HER2 scoring. The same principle—using AI to analyze motion-blurred or otherwise imperfect images—could be applied to other forms of tissue staining, biomarker analysis, or even entirely different clinical imaging modalities. As the researchers note, this work demonstrates a general framework for co-designing optical systems and deep learning algorithms to maximize diagnostic performance while minimizing hardware requirements.

Article: https://www.nature.com/articles/s41746-025-01882-x