RD-Light-Net: Light Weight Network for Retinal Detachment Classification through Fundus Images

Over a past few years, Deep Learning (DL) based techniques has becoming more popular and extensively used in the area of bio-medical imaging. Deep learning has enormous potential to accurately predict the diseases using fundus repositories. The prediction performance of DL is limited by training database size and requirement of computing memory. For addressing this, we presented a novel lightweight convolutional neural network based model for accurate prediction of Retinal detachment (RD) disease using retinal color fundus database. The proposed model provides a perfect solution for potential detection algorithms with limited datasets. Compared with other prior DL algorithms, our proposed work has less parameter which is more compatible with limited datasets. The proposed model has been evaluated with 1576 openly accessible retinal fundus color images. The proposed framework is able to recognize the RD and Non-RD automatically with statistical metrics namely sensitivity, specificity, accuracy, precision and F1-score respectively. The proposed model is employed and examines the various aspects for accurate recognition of RD using retinal fundus images that helps the medical practitioners in better decision making for the human eye.