An Early Experience Toward Developing Computer Aided Diagnosis for Gram-Stained Smears Images

Gram stained direct smears test is clinically useful in early identification of infections. Unfortunately, this practice is considered time consuming and labour intensive. Most existing effort in this area is to perform highmagnification analysis of images taken from manually selected areas. In this paper, we address the problem of the automatic selection of candidate areas (or patches) for subsequent high-magnification analysis. Specifically, we explore and study the possibility of selecting good working areas based on low-magnification images, where bacteria are likely to be found when viewed in high-magnification images. To this end, we develop an approach to classify the areas of interest according to the textural information of the image patch. We explore and study the efficacy of traditional textural features such as Histogram of Gradients, Local Binary Patterns, and 2 Dimensional Discrete Cosine Transform. Experiments show that the best variant method is able to select working areas where it is likely to find bacteria in high-powered objective images in a wide range of images.