Trends of Computer-aided Diagnosis in Estimating Tumor Malignancy

Abstract

This chapter encapsulates the trends of the computer-aided diagnosis (CAD) development for the estimation of tumor malignancy. Due to the high occurrence rate of cancer, more and more abnormal cell growth can appear at any organs in human body. To avoid unnecessary biopsy, the malignancy evaluation should be modeled and acquired via noninvasive examinations. CADs based on medical images are the effective tool for this purpose. Upon the developments of imaging modalities, more and more tumor characteristics can be observed and quantified. As an adjunct to mammography in breast cancer. Ultrasound techniques including B-mode, Doppler, elastogaphy are proposed to extract the physical, mechanical properties, and vessel distribution around tumors. Based on the artificial intelligence classifier, the complementary power of various features can be combined to deal with the heterogeneity of cancer types. These quantitative features can be classified into two categories:morphology and texture features. For ultrasound images, the echogenicities of different tissues are presented by gray-scale pixel values. The gray-scale intensities of region of interest can be analyzed without image segmentation or manual delineation and thus especially convenient for texture analysis. However, different settings or manufactures of ultrasound scanners usually result in different gray-scale distributions for the same tissue. The newly developed intensity-invariant texture features based on multi-resolution and orientations which successfully classified breast tumors into benign and malignant groups with accuracy higher than 80% are introduced as a solution. Finally, the use of CAD can be extended to precision medicine. Gene expression of everyone is quite different due to the inherent and environmental effects. In the past, the researches about gene are isolated to clinical examinations because the cost and the invasive procedures. With the development of image features extracted from medical images such as computed tomography, magnetic resonance imaging, ultrasound, and so on, the correlations between hundreds of quantitative image features and gene expressions are explored. The success of the researches can discover more tumor characteristics from image properties to gene expression for cancer stage estimation, survival rate, and many related issues. The role of CAD would be more important than before because of the computation of more data and more sophisticated techniques.