Developing a Mobile Application using Deep Learning for Cataract Classification

Tasnia Ishrat Khan; Fatima Ibrahim

Developing a Mobile Application using Deep Learning for Cataract Classification

dc.contributor.advisor	Mohammad Monirujjaman Khan
dc.contributor.author	Tasnia Ishrat Khan
dc.contributor.author	Fatima Ibrahim
dc.contributor.id	1911539642
dc.contributor.id	2121340642
dc.coverage.department	Electrical and Computer Engineering
dc.date.accessioned	2025-07-01
dc.date.accessioned	2025-07-01T05:02:35Z
dc.date.available	2025-07-01T05:02:35Z
dc.date.issued	2023
dc.description.abstract	One of the leading global causes of vision loss and blindness is the cataract. The percentage of blind people is around 50%. As a result, early cataract detection and prevention may limit vision loss and blindness. Contrary to cataract, artificial intelligence (AI) has made significant progress in the treatment of glaucoma, macular degeneration, diabetic retinopathy, corneal abnormalities, and age-related eye diseases. However, the vast majority of cataract detection algorithms in use are built using common machine learning techniques. On the other hand, manual extraction of retinal features is a laborious method that needs a skilled ophthalmologist. In order to detect cataracts, we have built the framework of an Android application. We then used algorithms to extract accuracy, graphs, trainable and untrainable parameters, and differentiation of cataract and non-cataract eye images from a gathered dataset. In order to identify the cataract using color fundus images, we presented the VGG19 (Visual Geometry Group), and digital image we presented Inception V3, which is a CNN (convolutional neural network) model. This will be incorporated into an Android application. The results of fundus image, the training procedure demonstrate that the model attained a flawless accuracy of 1.0000 on the training data for epochs 10 to 15. It scored an accuracy of 0.963 on the validation set, which is still quite high. With values ranging from 0.25 to 0.27, the validation loss was similarly largely consistent. The model is doing well and has mastered correctly classifying the photos. On the test data, the model produced a loss of 0.25735 and an accuracy of 0.9241. The result of the digital image, accuracy is 0.973 on the validation set, which is quite high and on the test data, the model produced a loss of 0.26753 and accuracy of 0.93491. The significance of these results is that the model performs effectively, can reliably categorize test photos with high accuracy, and will be trustworthy for patients to utilize.
dc.description.degree	Undergraduate
dc.identifier.cd	600000604
dc.identifier.print-thesis	To be assigned
dc.identifier.uri	https://repository.northsouth.edu/handle/123456789/1203
dc.language.iso	en
dc.publisher	North South University
dc.rights	© NSU Library
dc.title	Developing a Mobile Application using Deep Learning for Cataract Classification
dc.type	Thesis
oaire.citation.endPage	62
oaire.citation.startPage	1