Efficient DNN-Based Classification of Whole Slide Gram Stain Images for Microbiology

The interpretation of conventional glass Gram stain microscopy slides is both subjective and time consuming. The first step towards Digital Pathology is to convert Gram slides into Whole Slide Images (WSIs) - this image capture process itself is extremely challenging due to the need for x100 objectives with oil immersion for conventional microscopy. With high volume pathology laboratories, having an Artificial Intelligence (AI) system based on deep neural networks (DNNs) operating on WSIs could be extremely beneficial to alleviate problems faced by conventional pathology at scale. Such a system would ensure accuracy, reduce the workload of pathologists, and enhance both objectivity and efficiency. After reviewing the pathology literature, it is exceedingly rare to find methods or datasets relating to the very important Gram stain test compared to other pathology tests such as Breast cancer, Lymphoma and Colorectal cancer. This data scarcity has likely hindered research on Gram stain automation. This paper aims to use deep learning to classify Gram positive cocci bacteria subtypes, and to study the effect of downsampling, data augmentation, and image size on both classification accuracy and speed. Experiments were conducted on a novel dataset of three bacteria subtypes provided by Sullivan Nicolaides Pathology (SNP) comprising: Staphylococcus, Enterococcus and Streptococcus. The subimages are obtained from blood culture WSIs captured by the in-house SNP MicroLab using a x63 objective without coverslips or oil immersion. Our results show that a DNN-based classifier distinguishes between these bacteria subtypes with high classification accuracy.