Abstract
We developed a rapid scanning optical microscope, termed "BlurryScope", thatleverages continuous image acquisition and deep learning to provide acost-effective and compact solution for automated inspection and analysis oftissue sections. BlurryScope integrates specialized hardware with a neuralnetwork-based model to quickly process motion-blurred histological images andperform automated pathology classification. This device offers comparable speedto commercial digital pathology scanners, but at a significantly lower pricepoint and smaller size/weight, making it ideal for fast triaging in smallclinics, as well as for resource-limited settings. To demonstrate theproof-of-concept of BlurryScope, we implemented automated classification ofhuman epidermal growth factor receptor 2 (HER2) scores on immunohistochemically(IHC) stained breast tissue sections, achieving concordant results with thoseobtained from a high-end digital scanning microscope. We evaluated thisapproach by scanning HER2-stained tissue microarrays (TMAs) at a continuousspeed of 5 mm/s, which introduces bidirectional motion blur artifacts. Thesecompromised images were then used to train our network models. Using a test setof 284 unique patient cores, we achieved blind testing accuracies of 79.3% and89.7% for 4-class (0, 1+, 2+, 3+) and 2-class (0/1+ , 2+/3+) HER2 scoreclassification, respectively. BlurryScope automates the entire workflow, fromimage scanning to stitching and cropping of regions of interest, as well asHER2 score classification. We believe BlurryScope has the potential to enhancethe current pathology infrastructure in resource-scarce environments, savediagnostician time and bolster cancer identification and classification acrossvarious clinical environments.