Gram Stain Bacteria Classification Model Based On Convolution Neural Networks

Developing a Computer Aided Diagnostic (CAD) system for bacteria detection from direct smears images would help overcome the lack of expertise, laboratory set-ups and avoid the need for a culture process, thereby facilitating quick treatment. Gram stain analysis is suitable for low-resource settings. Gram stain sample before culture for the identification of pathogenic bacteria, indicative cells such as Neutrophils provide suggestive preliminary evidence of bacterial infection. Classification of bacteria from direct smear is challenging due to the bacteria size, presence of debris, and artifacts. The present work explores classification algorithms suitable for bacteria detection. 3 layer CNN method is designed suitable for low-resource settings. It is also compared with ResNet50 and Benchmark software Microsoft lobe. The proposed method gives promising result which could lead to early detection of bacteria