Classification of medical images based on deep stacked patched auto-encoders
- PDF / 2,107,837 Bytes
- 21 Pages / 439.642 x 666.49 pts Page_size
- 38 Downloads / 179 Views
Classification of medical images based on deep stacked patched auto-encoders Ramzi Ben Ali1 · Ridha Ejbali1 · Mourad Zaied2 Received: 10 April 2019 / Revised: 23 January 2020 / Accepted: 8 May 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The concept of artificial intelligence is not new. Without going into details of the evolution of artificial intelligence, we can confess that recent techniques of deep neural networks have considerably relaunched the trend with a significant advance namely the ability to automatically learn high-level concepts. However, a great step has been taken in deep learning to help researchers perform segmentation, feature extraction, classification and detection from raw medical images. This paper concerns the automatic classification of medical images with deep neural networks. We aimed at developing a system for automatic classification of medical images and detection of anomalies in order to provide a decision-making tool for the doctor. In this component we proposed a method for classifying medical images based on deep neural network using sparse coding and wavelet analysis. Serval real databases are used to test the proposed methods: MIAS and DDSM for mammogram images, LIDC-IDRI for lung images and dental dataset images. Classifications rates given by our approach show a clear improvement compared to those cited in this article. Keywords Images classification · Pattern recognition · Deep learning · Wavelet network · Deep neural wavelets network · Sparse auto-encoder
1 Introduction Automatic image classification is a pattern recognition application that automatically assigns a class to an image using a classification system. The classification of objects, Ramzi Ben Ali
[email protected] Ridha Ejbali ridha [email protected] Mourad Zaied [email protected] 1
Research Team in Intelligent Machines (RTIM), University of Gabes, ENIG, Zrig 6029, Gabes, Tunisia
2
RTIM, Gabes, Tunisia
Multimedia Tools and Applications
scenes, textures, face recognition, fingerprints and characters are among the common applications [17]. There are two main types of learning depending on the information available on the data to be classified: supervised and unsupervised learning. In the supervised approach, each image is associated with a label that describes its class of belonging. In the unsupervised approach (or clustering), the available data does not have labels; it is then up to the system to extract a membership rule from each image to a given group. In this presentation, we will deal only with the supervised approach and therefore for which label. An automatic image classification system consists of the following steps: the preprocessing step to “clean” the images; the feature extraction step to describe the relevant information contained in the image using discriminating operators or descriptors [30]; the learning step to build a decision boundary to identify the class of an image presented at the system entry. These three phases are essential in th
Data Loading...
