الفهرس | Only 14 pages are availabe for public view |
Abstract Chronic Kidney disease is one of the diseases that can cause death, in Egypt. Deaths caused by Kidney Diseases reached 3.41% of total deaths according to the World Health Organization. Hence, it is needed to develop and improve a tool to help the physicians to early detect kidney diseases. Medical Ultrasound is considered an effective, easy and inexpensive diagnostic tool for a variety of diseases compared to other modalities. Recently, its use has increased dramatically as a result of the development of new technologies that facilitate the production of high-quality images, in addition, these technologies made the ultrasound devices smaller and more affordable to the clinician. In the present work, an automated computer-aided system was designed to analyze and classify ultrasonic kidney images into five main classes: Cyst, Failure, Stone, Tumor and Normal. A comparative study was performed to analyze and discuss the performance of three feature extraction techniques in the classification of ultrasound kidney images. Different sets of image features were extracted using spatial-domain feature extraction that includes statistical features and transform-domain feature extraction using discrete wavelet transform (DWT) and wavelet packet transform (WPT). Principal component analysis was performed on wavelet analysis results for optimal feature selection. A neural network classifier was developed using the selected features in the training and testing phases. The results have shown that a correct classification rate of 98.6% has been achieved using wavelet-packet decomposition technique combined with PCA. It is concluded that the features derived from the wavelet-packet decomposition with PCA outperformed those derived from texture and Fourier transform using a neural network classifier. |