Affiliations: [a] Research and Development Centre, Bharathiar University, Coimbatore, Tamil Nadu, India
| [b] Department of Computer Science, C.B.M. College, Coimbatore, Tamil Nadu, India
Correspondence:
[*]
Corresponding author. S. Sadhana, Research Scholar, Research and Development Centre, Bharathiar University, Coimbatore, Tamil Nadu, India. E-mail: [email protected].
Abstract: Blindness is one of the serious issues in the present medical world scenario mainly caused by Diabetic Retinopathy (DR). It is a diabetes complication, that is produced due to the problems in retina blood vessel. For clinical treatment, it will be extremely helpful, if diabetic retinopathy is detected in early stages. In recent years, the manual detection of DR consumes more time and moreover, the detection of DR in early stages is still a challenging task. In order to avoid these issues, this research work focus on an automated as well as effective solution for detecting DR symptoms from retinal images and requires less time for accurate detection. A Novel histogram equalization technique is used for performing contrast enhancement and equalization in initial pre-processing stage. Then, from these pre-processed images, image patches are extracted regularly. Improved Discrete Curvelet Transform based Grey Level Co-occurrence Matrix (IDCT-GLCM) is used in second stage for extracting features. Then, extracted features are given to Classifier. At last, an Improved Alexnet model-based CNN (IAM-CNN) classification approach is used for diagnosing DR from digital fundus images. In terms of accuracy, specificity and sensitivity, effectiveness and efficiency of proposed method is shown by extensive simulation results.
Keywords: Improved alexnet model based CNN (IAM-CNN), improved discrete curvelet transform based grey level co-occurrence matrix (IDCT-GLCM), novel histogram equalization, diabetic retinopathy
