Automatic diagnosis of keratitis from low-quality slit-lamp images using feature vector quantization and self-attention mechanisms

Abstract

This paper proposes a novel method for the automatic diagnosis of keratitis using feature vector quantization and self-attention mechanisms (ADK_FVQSAM). First, high-level features are extracted using the DenseNet121 backbone network, followed by adaptive average pooling to scale the features to a fixed length. Subsequently, product quantization with residuals (PQR) is applied to convert continuous feature vectors into discrete features representations, preserving essential information insensitive to image quality variations. The quantized and original features are concatenated and fed into a self-attention mechanism to capture keratitis-related features. Finally, these enhanced features are classified through a fully connected layer. Experiments on clinical low-quality (LQ) images show that ADK_FVQSAM achieves accuracies of 87.7%, 81.9%, and 89.3% for keratitis, other corneal abnormalities, and normal corneas, respectively. Compared to DenseNet121, Swin transformer, and InceptionResNet, ADK_FVQSAM improves average accuracy by 3.1%, 11.3%, and 15.3%, respectively. These results demonstrate that ADK_FVQSAM significantly enhances the recognition performance of keratitis based on LQ slit-lamp images, offering a practical approach for clinical application.

Keywords

A

References

[1] TingD S J, HoC S, DeshmukhR, et al.. Infectious keratitis: an update on epidemiology, causative microorganisms, risk factors, and antimicrobial resistance. Eye, 2021, 35(4): 1084-1101. J]

[2] PINEDA R. World corneal blindness[J]. Foundations of corneal disease: past, present and future, 2020: 299–l305.

[3] LiZ, JiangJ, ChenK, et al.. Preventing corneal blindness caused by keratitis using artificial intelligence. Nature communications, 2021, 1213738. J]

[4] JiangJ, LiuW, PeiM, et al.. Automatic diagnosis of keratitis using object localization combined with cost-sensitive deep attention convolutional neural network. Journal of big data, 2023, 101121. J]

[5] XuY, KongM, XieW, et al.. Deep sequential feature learning in clinical image classification of infectious keratitis. Engineering, 2021, 7(7): 1002-1010. J]

[6] LiZ, JiangJ, QiangW, et al.. Comparison of deep learning systems and cornea specialists in detecting corneal diseases from low-quality images. Iscience, 2021, 2411103317. J]

[7] LiZ, JiangJ, ChenK, et al.. Development of a deep learning-based image quality control system to detect and filter out ineligible slit-lamp images: a multicenter study. Computer methods and programs in biomedicine, 2021, 203106048. J]

[8] CuiT, LinD, YuS, et al.. Deep learning performance of ultra-widefield fundus imaging for screening retinal lesions in rural locales. JAMA ophthalmology, 2023, 141(11): 1045-1051. J]

[9] KumarY, GuptaS. Deep transfer learning approaches to predict glaucoma, cataract, choroidal neovascularization, diabetic macular edema, drusen and healthy eyes: an experimental review. Archives of computational methods in engineering, 2023, 30(1): 521-541. J]

[10] CoanL J, WilliamsB M, AdithyaV K, et al.. Automatic detection of glaucoma via fundus imaging and artificial intelligence: a review. Survey of ophthalmology, 2023, 68(1): 17-41. J]

[11] MonemianM, RabbaniH. Exudate identification in retinal fundus images using precise textural verifications. Scientific reports, 2023, 1312824. J]

[12] JiangJ, ZhangY, XieH, et al.. A deep learning based fine-grained classification algorithm for grading of visual impairment in cataract patients. Optoelectronics letters, 2024, 20(1): 48-57. J]

[13] LiZ, QiangW, ChenH, et al.. Artificial intelligence to detect malignant eyelid tumors from photographic images. NPJ digital medicine, 2022, 5123. J]

[14] JiangJ, LiuH, HeL, et al.. HMADET: a hybrid model for automatic detection of eyelid tumors based on photographic images. BioMedical engineering online, 2024, 23125. J]

[15] ChenY, GuanT, WangC. Approximate nearest neighbor search by residual vector quantization. Sensors, 2010, 10(12): 11259-11273. J]

[16] TolstikhinI O, HoulsbyN, KolesnikovA, et al.. MLP-mixer: an all-MLP architecture for vision. Advances in neural information processing systems, 2021, 34: 24261-24272[J]

[17] BuslaevA, IglovikovV I, KhvedchenyaE, et al.. Albumentations: fast and flexible image augmentations. Information, 2020, 112125. J]

[18] SelvarajuR R, CogswellM, DasA, et al.. Grad-CAM: visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE International Conference on Computer Vision, December 22–29, 2017, Venice, Italy, 2017, New York. IEEE. 618626[C]

RIGHTS & PERMISSIONS

Tianjin University of Technology