A PROFESSIONAL TOOL FOR KNOWLEDGE EXTRACTION FROM DICOM IMAGES

Abstract

automatic analysis on medical data collected. Techniques from computer vision domain are being adopted and customized for the analysis of medical images. Most of the tools available perform well for a certain class of images. The performance of these tools is usually expressed as a relative metric to the performance of other tools for the same task. The broad range of algorithms available necessitates performance analysis on a common data set. In the current work the issue is addressed by providing a professional tool that allows techniques developed for medical image analysis to be tested on synthetic data sets. Users are allowed to generate images of both regular and irregular structures with varying level of noise in different components of the image. By testing the performance of the algorithms on known shapes with known noise levels the system is provided with the knowledge of which algorithm performs well for a particular class of images. DICOM being the most prevalent format for representing medical images, the tool provides synthetic images in DICOM format. To prove the effectiveness of the tool we study the performance of two common segmentation algorithms with various synthetic images.