e-space
Manchester Metropolitan University's Research Repository

    Comparative study of pattern recognition methods for predicting glaucoma diagnosis

    Williams, L, Waqar, S, Sherman, T and Masala, G ORCID logoORCID: https://orcid.org/0000-0001-6734-9424 (2020) Comparative study of pattern recognition methods for predicting glaucoma diagnosis. In: Innovation in Medicine and Healthcare. Smart Innovation, Systems and Technologies. Springer, pp. 93-103. ISBN 978-981-15-5852-8

    [img]
    Preview
    Accepted Version
    Download (240kB) | Preview

    Abstract

    © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2020. Glaucoma is the second leading cause of blindness globally; it is characterised by degeneration of the optic nerve with particular patterns of corresponding defects in the visual field. Aiding doctors in early diagnosis and detection of progression is crucial, as glaucoma is asymptomatic in nature. Furthermore there are good therapeutic results in early cases before irreversible visual loss occurs. Thus it is of great importance to find automated methods to discriminate glaucomatous diseases giving insight to doctors. In order to develop a Computer-Aided Diagnosis system (CAD), we realised an extensive competitive study of pattern recognition methods should be undertaken. A range of methods has been evaluated including the use of Deep Neural Networks (DNN), Support Vector Machines (SVM), Decision Trees (DT), and K-Nearest Neighbours (KNN) for diagnosing glaucoma. Using a range of classification techniques, this paper aims to diagnose glaucomatous diseases. Results have been produced with data comprising of Visual Field and OCT Disc readings from anonymous patients with and without glaucoma. Multiple systems are proposed that can predict diagnosis for ocular hypertension, primary open-angle glaucoma, normal tension glaucoma, and healthy patients with a reasonable confidence. Best performance has been obtained from voting classier comprised of SVM and KNN at 0.87 (AUC) and DNN at 0.87 (AUC) which possibly could be used as an automatic diagnosis aid in order to streamline the diagnosis of glaucoma for complex cases or flagging of urgent cases.

    Impact and Reach

    Statistics

    Activity Overview
    6 month trend
    0Downloads
    6 month trend
    0Hits

    Additional statistics for this dataset are available via IRStats2.

    Altmetric

    Repository staff only

    Edit record Edit record