Diagnosis of Focal Liver Diseases Based on Deep Learning Technique for Ultrasound Images

• Published in: Arabian journal for science and engineering (2011) Vol. 42; no. 8; pp. 3127 - 3140
• Main Authors: Hassan, Tarek M., Elmogy, Mohammed, Sallam, El-Sayed
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2017; Springer Nature B.V
• Subjects: Accuracy; Bayesian analysis; CAI; Classification; Clustering; Color; Computer assisted instruction; Deep learning; Diagnosis; Diseases; Engineering; Feature extraction; Fuzzy sets; Humanities and Social Sciences; Image classification; Image segmentation; Lesions; Liver; Liver diseases; Mammography; Medical imaging; multidisciplinary; Pixels; Preprocessing; Research Article - Computer Engineering and Computer Science; Science; State of the art; Ultrasonic testing; Ultrasound; Deep learning; Softmax classifier; Stacked sparse auto-encoders; Computer-aided diagnosis; Focal liver diseases diagnosis
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-016-2387-9

In the last three decades, there is considerable interest in computer-aided diagnosis systems for dealing with different diseases. Recently, these computer-aided diagnosis systems use deep learning architectures for analysis and classification of medical images. The previous techniques consider the hand-designed feature extraction approaches that depend on low-level image features, such as edges, color, and texture. Unlike these techniques, in this paper, a feature representation with a stacked sparse auto-encoder that is based on deep learning technology is proposed. The stacked sparse auto-encoder is trained in an unsupervised way. It learns the high-level features of the input pixels of unlabeled images that differentiate among different images that contain the various focal liver diseases. The proposed system consists of the preprocessing stage followed by the segmentation of the liver lesions using the level set method and Fuzzy c-means clustering algorithm. The stacked sparse auto-encoder extracts the high-level features representation from pixels of the segmented images, which are considered as the inputs for the classifier. Finally, a softmax layer classifies the different focal liver diseases by selecting the highest probabilities of each class. Using our proposed system, we have got an overall classification accuracy of 97.2%. Our proposed system is compared with three state-of-the-art techniques which are multi-support vector machine, K-Nearest Neighbor, and Naive Bayes. The experimental results show that the accuracy of classification of our proposed system outperforms the three state-of-the-art techniques.