Computer-aided diagnosis of cataract using deep transfer learning. (August 2019)

• Record Type: Journal Article
• Authors: Pratap, Turimerla; Kokil, Priyanka
• Is Part Of: Biomedical signal processing and control. Volume 53(2019)
• Publication Date: 2019-08
• Subjects: Computer-aided diagnosis -- Pre-trained convolutional neural networks -- Feature extraction -- Support vector machine classifier -- Transfer learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.04.010
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A computer-aided transfer learning-based automatic cataract detection method is proposed. Automatic classification is done via pre-trained CNN for deep transfer learning. Image quality assessment module is implemented to discard poor quality images. The proposed method outperforms with the existing methods in evaluating the cataract. Proposed method also eliminates the burden of training a CNN from scratch. Abstract: Cataract is a leading eye disease across the world. If cataract is not diagnosed in earlier stage, then it may lead to blindness. Earlier detection is the best way to control the risk and to avoid painful surgery. Thus, this paper mainly focuses on cataract detection from fundus retinal images. A computer-aided automatic cataract detection method is proposed to detect various stages of the cataract such as normal, mild, moderate, and severe from the fundus images. The proposed method uses the pre-trained convolutional neural network (CNN) for the transfer learning to carry out automatic cataract classification. Then pre-trained CNN model is used for the feature extraction and the extracted features are then applied to a support vector machine (SVM) classifier. The fundus cataract images are collected from the various open access datasets and labelled into four stages with the help of ophthalmologic experts. The four stage classification accuracy obtained is 92.91%. Since, the image quality is important in CNN, an image quality selection module is incorporated to decide the quality of fundus image for diagnosis. The revaluation of results based on the quality of fundus images is also presented. Based on the results, the proposed method proved to be an efficient method that uses pre-trained CNN as transfer learning for the classification of the cataract.