IRIS-CDSS: An Explainable Multimodal Deep Learning System with CBAM Attention for Longitudinal Cataract Progression and Surgical Decision Support

In this study, we have formulated and evaluated IRIS-CDSS, an auditable, weakly-supervised deep learning framework for the automated grading, spatial localization, longitudinal progression tracking, and surgical scheduling of cataracts from slit-lamp biomicroscopy photographs. Utilizing an EfficientNet-B0 backbone augmented with a Convolutional Block Attention Module (CBAM), the framework achieves an overall macro-averaged accuracy of 95.60%, a sensitivity of 95.84%, and an AUC-ROC of 0.9922 across the clinical benchmark test cohort. Future research will focus on extending the architectural framework to smartphone-based anterior segment imaging adapters to facilitate tele-ophthalmology in remote primary healthcare centers, integrating automated pupil segmentation with lightweight edge models, and deploying the platform across multi-center clinical trials to support community-level preventable blindness eradication initiatives.

[1]           World Health Organization, "World report on vision," WHO Guidelines Approved by the Guidelines Review Committee, Geneva, Switzerland: World Health Organization, 2019. [2]            L. T. Chylack, J. K. Wolfe, D. M. Singer, M. C. Leske, M. A. Bullimore, et al., "The Lens Opacities Classification System III," Archives of Ophthalmology, vol. 111, no. 6, pp. 831–836, Jun. 1993. [3]            J. Nayak, S. Saxena, S. K. Lakshminarayanan, and S. V. Sree, "Automated classification of cataract stages using deep learning methods," Computers in Biology and Medicine, vol. 132, p. 104334, May 2021. [4]            E. Long, H. Lin, Z. Liu, X. Wu, L. Wang, et al., "An artificial intelligence platform for the multihospital collaborative management of congenital cataracts," Nature Biomedical Engineering, vol. 1, no. 2, pp. 1–8, Feb. 2017. [5]           X. Zhang, G. Xiao, Y. Zou, Y. Lin, and Y. Ke, "Automated nuclear cataract grading based on anterior segment optical coherence tomography," IEEE Transactions on Medical Imaging, vol. 39, no. 6, pp. 2024–2034, Jun. 2020. [6]           A. Pratap, R. K. Patra, P. P. Roy, and B. G. Dogra, "Mobile-based screening of lens opacities using lightweight convolutional neural networks," IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 3, pp. 1120–1131, Mar. 2022. [7]            S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, "CBAM: Convolutional Block Attention Module," in Proc. European Conference on Computer Vision (ECCV), Munich, Germany: Springer, 2018, pp. 3–19. [8]            R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad-CAM: Visual explanations from deep networks via gradient-based localization," in Proc. IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017, pp. 618–626. [9]            Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, "Image quality assessment: from error visibility to structural similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, Apr. 2004. [10]            M. Tan and Q. V. Le, "EfficientNet: Rethinking model scaling for convolutional neural networks," in Proc. 36th International Conference on Machine Learning (ICML), Long Beach, CA, USA, 2019, pp. 6105–6114.