A novel multimodality based dual fusion integrated approach for efficient and early prediction of glaucoma. (March 2022)

• Record Type: Journal Article
• Authors: Singh, Law Kumar; Khanna, Munish; Pooja,
• Is Part Of: Biomedical signal processing and control. Volume 73(2022)
• Publication Date: 2022-03
• Subjects: Multimodal -- Glaucoma prediction -- Fundus images -- Deep Learning -- Machine Learning; 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.2021.103468
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel multimodality based approach for efficient Glaucoma prediction is proposed. Early fusion and late fusion both are implemented in this work. Machine Learning and Deep learning are implemented using feature level fusion and image level fusion respectively. Approach is tested on three benchmark datasets and four combinations of these datasets. Accuracy up to 95.56% is achieved through this approach. Abstract: As there is currently no exact treatment for glaucoma, early detection and diagnosis are essential to reduce the risk of this infection. In recent years, Machine learning and deep learning has significantly improved prediction and classification of human diseases. We are the first to offer a new multimodal approach for glaucoma prediction in this article. We shortlisted three public datasets and in totality we tested seven combinations of these datasets. Initially, we created five multimodal representations of each publicly accessible benchmark dataset. In the first vertical, we extracted 36 critical features from each multimodal of a particular dataset. These extracted features are subsequently fused (referred to as early fusion) to create each dataset's 180 features. These 180 features are ranked using random forest. The top 50% of the features are retrieved to create a feature vector. This feature vector is fed into different machine learning classifiers and their ensemble model for classification purposes. In the second vertical, we worked at the picture level where we send images from each dataset's five multimodal dimensions to two deep learning methods for classification purposes. For each of the seven experiments conducted in this study we obtain several sets of findings. These categorization findings are combined (referred to as late fusion) and submitted to professional ophthalmologists who make the final determination based on their judgments. As a consequence of the proposed approach, we now have a computerized glaucoma diagnostic system with remarkable results (accuracy upto 95.56%).