Predicting spatial displacement based on intraocular image design using convolution neural network - preliminary findings

• Main Authors: Mohd Tamrin, Mohd Izzuddin, Che Azemin, Mohd Zulfaezal, Md Noor Rudin, Noor Fawazi, Mohd Salleh, Mohd Hazimin, Hilmi, Mohd Radzi, Alwan, Ali Amer, Shah, Asadullah
• Format: Article
• Language: English
• Published: IIUM Press 2021
• Subjects: TK7885 Computer engineering
• Online Access: http://irep.iium.edu.my/89550/7/89550_Predicting%20spatial%20displacement%20based%20on%20intraocular%20image%20design%20using%20convolution%20neural%20network%20-%20preliminary%20findings.pdf; http://irep.iium.edu.my/89550/; https://journals.iium.edu.my/kict/index.php/jisdt/article/view/206/132

The main global cause for blindness is due to cataract. The common treatment for cataract is to have the cloudy natural lens removed and replaced with an artificial intraocular lens (IOL). Success in the post cataract surgery depends on the design and quality of the IOL implanted on the eye. ISO11979-3 is the standard adhered to by many lens manufacturers, to test the mechanical stability of the lenses that they produced. This compression test experiments on the lab are very costly and time consuming. Alternatively, we propose to use the convolution neural network (CNN) to predict the spatial displacement response based on the intraocular image designs. Due to limited number of images in the datasets, data augmentation was performed to transform these images and increase the sample size to 240. On top of this, the ResNet-50 deep learning network architecture was utilized to transfer the learning done on over millions of images. The final RMSE value for the training set, validation set and testing set were at 0.47mm, 2.93mm and 2.92mm respectively. The model predictabillity is well within the range recommended by the standard between 0.15 to 1.98 mm.