Multi-class Classification of Retinal Eye Diseases from Ophthalmoscopy Images Using Transfer Learning-Based Vision Transformers

Deleted Journal, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 27, 2025

This study explores a transfer learning approach with vision transformers (ViTs) and convolutional neural networks (CNNs) for classifying retinal diseases, specifically diabetic retinopathy, glaucoma, cataracts, from ophthalmoscopy images. Using balanced subset of 4217 images ophthalmology-specific pretrained ViT backbones, this method demonstrates significant improvements in classification accuracy, offering potential broader applications medical imaging. Glaucoma, cataracts are common eye diseases that can cause loss if not treated. These must be identified the early stages to prevent damage progression. paper focuses on accurate identification analysis disparate including using Deep (DL) has been widely used image recognition detection treatment diseases. In study, ResNet50, DenseNet121, Inception-ResNetV2, six variations employed, their performance diagnosing such as retinopathy is evaluated. particular, article uses transformer model an automated diagnose highlighting accuracy pre-trained deep (DTL) structures. The updated ViT#5 augmented-regularized (AugReg ViT-L/16_224) rate 0.00002 outperforms state-of-the-art techniques, obtaining data-based score 98.1% publicly accessible dataset, which includes most categories, other convolutional-based models terms precision, recall, F1 score. research contributes significantly analysis, demonstrating AI enhancing precision disease diagnoses advocating integration artificial intelligence diagnostics.

Language: Английский

---

VNLU-Net: Visual Network with Lightweight Union-net for Acute Myeloid Leukemia Detection on Heterogeneous Dataset

Biomedical Signal Processing and Control, Journal Year: 2025, Volume and Issue: 107, P. 107840 - 107840

Published: March 29, 2025

Language: Английский

---

Enhanced ResNet50 for Diabetic Retinopathy Classification: External Attention and Modified Residual Branch

Mathematics, Journal Year: 2025, Volume and Issue: 13(10), P. 1557 - 1557

Published: May 9, 2025

One of the common microvascular complications in diabetic patients is retinopathy (DR), which primarily impacts retinal blood vessels. As course diabetes progresses, incidence DR gradually increases, and, serious situations, it can cause vision loss and even blindness. Diagnosing early essential to mitigate its consequences, deep learning models provide an effective approach. In this study, we propose improved ResNet50 model, replaces 3 × convolution residual structure by introducing external attention mechanism, improves model’s awareness global information allows model grasp characteristics input data more thoroughly. addition, multiscale added branch, further ability extract local features features, processing accuracy image details. Sophia optimizer introduced replace traditional Adam optimizer, optimizes classification performance model. 3662 images from Kaggle open dataset were used generate 20,184 for training after preprocessing augmentation. Experimental results show that achieves a 96.68% on validation set, 4.36% higher than original architecture, Kappa value increased 5.45%. These improvements contribute diagnosis decrease likelihood blindness among patients.

Language: Английский

---

Image-level multi-label retinal disease classification with a novel classification head

Computers & Electrical Engineering, Journal Year: 2025, Volume and Issue: 124, P. 110410 - 110410

Published: May 1, 2025

Language: Английский

---

Fundus-DANet: Dilated Convolution and Fusion Attention Mechanism for Multilabel Retinal Fundus Image Classification

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(18), P. 8446 - 8446

Published: Sept. 19, 2024

The difficulty of classifying retinal fundus images with one or more illnesses present missing is known as multi-lesion classification. challenges faced by current approaches include the inability to extract comparable morphological features from different lesions and resolve issue same lesion, which presents significant feature variances due grading disparities. This paper proposes a multi-disease recognition network model, Fundus-DANet, based on dilated convolution. It has two sub-modules address aforementioned issues: interclass learning module (ILM) dilated-convolution convolutional block attention (DA-CBAM). DA-CBAM uses (CBAM) convolution merge multiscale information images. ILM channel mechanism map lower dimensions, facilitating exploring latent relationships between various categories. results demonstrate that this model outperforms previous models in multilocular OIA-ODIR dataset 93% accuracy.

Language: Английский

---