Computer vision method for biomedical image analysis

Abstract

The modern technology utilizes all types of images as a source of interpretation and analysis. The proliferation of pictorial data has created the need for a vision based automation, that can rapidly, accurately, and cost effectively extract the useful information contained in images. This paper presents the development of software to process the medical images and extracts the necessary features such as area, perimeter, colour information, irregularity index, chromotocity, cytoplasmic inclusions, and variegation of colouring etc, required to diagnose various diseases. The software has been developed in Turbo C++ on a DOS-platform. It converts and resizes the scanned images into BMP format. The software is menu driven and performs the histogram-based segmentation of the image and boundary detection by applying Quick or sobel mask as desired. The facility of applying the advanced operations like Gaussian and Mexican Hat filters, useful in tracking the guide wire in angiographic images, are also incorporated. The developed software has been applied on various images of blood samples to classify the leukocytes, X-ray images to detect the boundary of the image, angiographic images to track the guide wire, and digital images of skin lesions (tumors) to detect skin malignancy and the results found to be satisfactory.