BlurryScope enables compact, cost-effective scanning microscopy for HER2 scoring using deep learning on blurry images

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Digital Health

Background:

• Automated analysis of tissue sections is crucial for accurate disease diagnosis.
• Current digital pathology scanners are often expensive, large, and inaccessible for many labs.
• There is a need for cost-effective, compact, and automated solutions for tissue imaging and analysis.

Purpose of the Study:

• To develop and validate a novel, rapid scanning optical microscope, BlurryScope, for automated tissue section analysis.
• To assess the performance of BlurryScope in classifying human epidermal growth factor receptor 2 (HER2) scores in breast cancer tissue.
• To demonstrate BlurryScope as a viable, low-cost alternative to commercial digital pathology scanners.

Main Methods:

• Development of BlurryScope, a rapid scanning optical microscope utilizing continuous image acquisition and deep learning algorithms.
• Implementation of automated image stitching, cropping, and HER2 score classification workflow.
• Comparative analysis of BlurryScope's HER2 classification accuracy against a high-end digital scanning microscope.

Main Results:

• BlurryScope achieves comparable speed to commercial digital pathology scanners at a significantly lower cost and smaller footprint.
• Automated HER2 classification on motion-blurred immunohistochemically stained breast tissue sections yielded high accuracy.
• Achieved 79.3% accuracy for 4-class (0, 1+, 2+, 3+) and 89.7% accuracy for 2-class (0/1+, 2+/3+) HER2 classification on 284 patient cores.

Conclusions:

• BlurryScope offers a cost-effective, compact, and automated solution for digital pathology.
• The system demonstrates high accuracy in HER2 scoring, comparable to high-end scanners.
• BlurryScope has the potential to improve accessibility and efficiency in tissue analysis and cancer diagnostics.