ODMNet: Automated Glaucoma Detection and Classification Model Using Heuristically-Aided Optimized DenseNet and MobileNet Transfer Learning

Abstract

In various existing works, glaucoma detection is not predicted accurately, which may lead to irreversible vision loss. A new framework is designed for detecting glaucoma by transfer learning approach. In the initial stage, the source images are gathered from standard datasets. After collecting the raw images, it is fed for image enhancement, performed through the Retinex approach. Further, the segmentation process is carried out by Modified DeepLabV3, where the significant Regions of Interest (ROI) are extracted by enhanced images and segmented the abnormalities. To meet the optimal value, the parameters in DeepLabV3 are tuned optimally by the Improved Rain Optimization Algorithm (IROA). Once the image is segmented, it is subjected to the detection or classification task, where the glaucoma is effectively classified by the hybrid learning method called Optimized DenseNet and MobileNet Transfer Learning (ODMNet) that is constructed with Densely Connected Convolutional Networks (DenseNet) and MobileNet, where the layers are optimized by IROA approach. Finally, the performance is assessed with the assistance of diverse metrics. In experimental analysis, the accuracy and F1-score of the designed method attain 96% and 93%. The recommended detection model achieves higher detection performance in telemedicine and healthcare applications.

Keywords:

• Abnormality segmentation
• automated glaucoma detection
• improved rain optimization algorithm
• modified DeepLabV3
• optimized DenseNet and MobileNet transfer learning
• retinex-based image enhancement