Automated detection of glaucoma using elongated quinary patterns technique with optical coherence tomography angiogram images. (August 2021)

• Record Type: Journal Article
• Authors: Chan, Yam Meng; Ng, E.Y.K.; Jahmunah, V; Koh, Joel En Wei; Oh, Shu Lih; Han, Wei Shan; Yip, Leonard Wei Leon; Acharya, U Rajendra
• Is Part Of: Biomedical signal processing and control. Volume 69(2021)
• Publication Date: 2021-08
• Subjects: Glaucoma -- Optical coherence tomography angiography -- Elongated quinary patterns -- Ensemble classifier -- Ten-fold validation -- Image features; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102895
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Classification of normal and glaucoma OCT images is proposed. Elongated quinary pattern and ensemble classifier are used. Various texture features are extracted. Ten-fold cross validation is employed to develop the system. Obtained high classification accuracy of 95.1%. Abstract: Glaucoma is the second most common cause of blindness worldwide after cataracts. It presents a great health concern as it is usually undetectable during the early stages without regular screening. Noticeable symptoms of glaucoma may only appear at a later stage. The eye disease progresses over time without treatment. Clinicians are specially trained to identify and diagnose glaucoma. However, reasons such as fatigue and observer errors may impair the clinician's judgement. Hence, a trained computer-aided diagnosis system is necessary to prevent such issues. Optical coherence tomography angiography (OCTA) images were used to detect glaucoma. In this work, we have used elongated quinary patterns (EQP) technique to obtain multi-gradient magnitudes and angles. Various texture features are extracted from the various levels of gradients and angles of EQP images. Optimal features selected using Student's t -test are fed to an ensemble classifier and 10-fold cross validation strategy is employed in which adaptive synthetic (ADASYN) is applied to reduce the bias. In this work, we have obtained an accuracy of 95.1% for the detection of left eye (OS) disc centered OCTA images. This developed system is available for further evaluation using more images.