Abstract: We offer a novel Convolutional Neural Network (CNN) method designed exclusively for detecting diabetic retinopathy (DR) in visual pictures. The existence of DR at an early stage has an impact on the effectiveness of treatment. Ophthalmologists commonly physically identify DR in retinal pictures; our objective is to do it correctly. To improve accuracy and avoid overfitting, we develop a specialised CNN architecture, clean the data, and employ a customised dataset. We increase actuation capabilities and hyperparameters through meticulous design. Our technique beats earlier frameworks in terms of accuracy, review, F1-score, and ROC-AUC, according to our research. By seeing the CNN highlight maps, one may have a better grasp of diagnosis. Our findings show that updated deep learning models might be used in restorative imaging to deliver rapid and accurate DR diagnosis, save medical staff workload, and possibly even safeguard patients' eyesight.
Keywords: Diabetic Retinopathy, CNN Architecture, Image Classification, Deep Learning, Customized Model, Medical Imaging
|
DOI:
10.17148/IJARCCE.2024.13437
