Matching fusion framework on multi-modal data for glaucoma severity diagnosis

• Published in: Computers & electrical engineering 2025-04, Vol.123, p.109982, Article 109982
• Main Authors: Yi, Sanli, Feng, Xueli
• Format: Article
• Language: English
• Subjects: CNNs; Matching fusion; Multi-modal data; Severity glaucoma diagnosis
• ISSN: 0045-7906
• DOI: 10.1016/j.compeleceng.2024.109982

In response to the challenge of limited medical datasets hindering the effective extraction of crucial features for glaucoma severity diagnosis, we propose a novel matching fusion framework based on multi-modal data. In the framework, firstly, a matching fusion method that matches different networks for different types of datasets is proposed. Secondly, two fusion strategies based on the matching fusion method are designed, multi-data fusion strategy and multi-model fusion strategy. Thirdly, four fusion methods based on three glaucoma data types are designed. Lastly, a new fusion classifier for the obtained fusion feature map is proposed,which is mainly composed of a fusion feature convolution block and a multi-layer perceptron block. To verify the framework, we conducted glaucoma classification experiments on the datasets collected from the First Affiliated Hospital of Kunming Medical University, achieving an accuracy of 0.965, an F1 score of 0.966, and an AUC of 0.997. The results demonstrated that our framework outperformed state-of-the-art methods.