Enhancement and Characterization of Cancerous Tissue Using Computed Tomography Images

Abstract

Computed tomography (CT) scan is most common imaging modality for diagnosis of various types of the diseases. The diagnosis of disease using CT images is done in two ways, one by visual inspection and other by computer aided diagnosis. The high quality of visible details is required in CT images for visual inspection by radiologists. For high quality CT images enhancement is required. The information present in the CT scan of tissues exists in the texture part of the CT images. CT images have good structural details but textural details need enhancement. The images obtained after enhancement assists radiologist to diagnose the disease easily and with less fatigue. The present work proposed two enhancement methods for CT images. First method is based on the transformation function of histogram equalization and it is modified by adding constrained variable offset to preserve the over enhancement produced by histogram equalization. This method improves the visual quality of the images and tumor is clearly visible in comparison to the original image. The second method is modified object based contrast stretching. In this method, first image is segmented using watershed method then it is divided into parts, object approximation image (OAI) and object error image (OEI). OEI is enhanced using greedy iterative stretching, and then mean adjustments are made to enhanced OEI and original OAI. To obtain final enhanced results both the images are added. The second method enhances the textural details of the CT images. Both methods are applied on 197 liver CT images database. The objective evaluation of both methods is done by quantitative performance measures. The evaluation shows that modified object based contrast stretching method (MOBCS) is better in comparison to histogram equalization with constrained variable offset (HECVO). To verify the textural enhancement of both method texture feature extraction is done using various feature extraction methods. Then feature selection is done using box plot study and sequential feature selection method. Feature selection show that HECVO method is not suitable for the texture feature analysis, so further work is done using results of MOBCS method. Classification of normal and abnormal liver tissue is done using support vector machine. The experimental results show that MOBCS performs better than original image in classifying the normal and abnormal tissue of the CT image.