D. Smutek (Czech Republic, Japan), A. Shimizu (Japan), L. Tesar (Japan, Czech Republic), H. Kobatake, S. Nawano (Japan), and P. Maruna (Czech Republic)
Image processing, Computed tomography, Ultrasound imaging, Liver lesions, Thyroid disorders
A computed-aided diagnostic system (CAD) for remote automatic classification of CT and B-mode ultrasound images is described. The system is capable of discriminating among focal liver lesions and chronic thyroid inflammatory diseases. The application is based on texture analysis. Two types of texture features are used in the system: 22 first-order features computed from the original gray levels and four different gray-level transformations of an image and 108 second-order features (computed from co-occurrence matrices) which capture the spatial organization of texture primitives. The classification of images is performed by network of Bayes classifiers with majority voting. The classifier was trained by Gaussian mixture model method and classifies images according to their texture feature values. In testing phase the system achieved 100% classification success rate when using four principal descriptive features. The results are sufficiently consistent under small changes in image tool setting and scan type. The implementation of the remote CAD system is promising for automatic classification.
Important Links:
Go Back
