Abstract—Glistening Quantification is essential to assess the quality of the lens and to assign appropriate treatment after the cataract surgery. Glistening increase forward light scattering and affect human vision. Glistenings are observed and graded in intraocular lens by clinicians through slit-lamp. Several studies report to grading by number of glistenings present in Intraocular lens (IOL). Since glistenings are small water inclusions in IOL and can be observed with different sizes. Manual Clinical grading is time consuming to finish each IOL and moreover, inaccurate. Therefore, one of the methods to quantify glistenings in IOL is required for both clinicians and patients. In this paper, Deep Learning approach based on convolution neural network is trained to quantify glistening in IOL. The result shows the proposed method can automatically classify glistenings in vitro IOL image and predict the probabilities of glistening in IOL. Index Terms— glistening detection, quantification, classification, deep learning I. INTRODUCTION LISTENING are fluid-filled microvacuoles in Intraocular lens which can be observed after cataract surgery. [1] [2]The lens is made up of water and protein molecules. In fact, the cataract is associated with aging process because the proteins of the lens start to clump together and cloud the lens is called a cataract. [3] [4] [5] [6] [7] However, this cloudy lens affect vision and therefore, Intraocular lens, medical devices that are implanted inside the eye and removed the natural lens during cataract surgery. [8] Glistenings are observed after 2 or 3 years of implanted surgery and occurred 90% in AcrySof® lenses. [9] Manuscript received March 20, 2018; revised April 04, 2018. This work was supported in part by Thammasat University, Japan Advanced Institute of Science and Technology and City University of London. Kay Thwe Min Han is with dual-degree program from SIIT, Thammasat University, Thailand and Japan Advanced Institute of Science and Technology. She is now with the Department of Information, Computer and Communication Technology (ICT), SIIT, Thammasat University, 131 Moo 5, Tiwanont Road, Bangkadi, Muang, Pathumthani, 12000, Thailand (phone: +66 (0) 2501 3505-20; fax: +66 (0) 2501 3524; e-mail: kthweminhan@ jasit.ac.jp). Bunyarit Uyyanonvara is with SIIT, Thammasat University, Thailand. He is now with the Department of Information, Computer and Communication Technology (ICT), 131 Moo 5, Tiwanont Road, Bangkadi, Muang, Pathumthani, 12000, Thailand (email: bunyarit@siit.tu.ac.th). Kotani Kazunori is with the School of Information Science, Entertainment Technology Area, Japan Advanced Institute of Science and Technology, 1-1 Asahidai, Nomi, Ishikawa 923-1292 Japan. (e-mail: ikko@jaist.ac.jp). Chris Hull is with the School of Health Sciences, Division of Optometry & Visual Sciences, City, University of London, Northampton Square, London EC1V 0HB,United Kingdom. (e-mail: c.c.hull@city.ac.uk). . The formation of glistenings is observed in the temperature changes of the lens from room temperature to body temperature in which the material absorbed water allowing to collect into voids within IOL over an extended period and form fluid-filled bubbles. [10] These bubbles appear to sparkle under slit-lamp examination and glistenings can affect the light scatter in IOL. Therefore, glistening grading is assessed by clinicians observing in vitro through a slit- lamp. [11] Yet, clinical grading is time consuming and the manual grading results are challenging due to distribution, number and size of glistening in IOL. Glistening detection with image processing techniques has been observed from the previous studies. The algorithm based on blob detection and watershed algorithm can detect glistenings semi-automatically. [12] For those methods, the foreground extraction process is important to be accurate and the background images are required to be stable. In this paper, deep learning is applied to perform an automated glistening quantification using vitro images. The proposed methodology aims to differentiate between different number of glistenings lens and non-glistening lens in IOL. As a result, the doctors can proceed any appropriate treatment based on the occurrence of glistening in IOL. Deep learning is used from the initial step of training the raw input images to the final quantification. Thus, overview of deep learning is described in section 2 and other sections are structured as follows: section 3 describes the methodology of this research and section 4 presents the result and discussion are followed as section 5. II. DEEP LEARNING Deep learning (DL) is the subfield of machine learning and increase number of interesting in computer vision, speech recognition, natural language processing, object detection, and audio recognition. [13] Nowadays, many researchers in medical imaging contribute in the field of deep learning and become one of the methodologies for analyzing medical images. In addition, deep learning can provide the optimal solutions with good accuracy for medical imaging and is anticipated for future applications in health sector. [14]One of the most popular uses of the ML algorithms is probably quantification. Although various DL architectures are emerged in recent years, Convolutional Neural Networks (CNN) is a common used architecture for complex operations which are required to use convolution filters. Neural Networks are essentially mathematical models to solve an optimization problem. [15] Typically, neural networks are biological inspired paradigm that enables computer to learn from data. Therefore, they are made of neurons, the basic computation unit of neural networks which have learnable weights and biases. [16] Each Kay Thwe Min Han, Bunyarit Uyyanonvara, Kotani Kazunori, Chris Hull Deep Learning for Glistening Quantification in Intraocular Lens G Proceedings of the World Congress on Engineering 2018 Vol I WCE 2018, July 4-6, 2018, London, U.K. ISBN: 978-988-14047-9-4 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) WCE 2018