Multiscale attention-over-attention network for retinal disease recognition in OCT radiology images DOI Creative Commons
Abdulmajeed M. Alenezi,

Daniyah A. Aloqalaa,

Sushil Kumar Singh

et al.

Frontiers in Medicine, Journal Year: 2024, Volume and Issue: 11

Published: Nov. 8, 2024

Retinal disease recognition using Optical Coherence Tomography (OCT) images plays a pivotal role in the early diagnosis and treatment of conditions. However, previous attempts relied on extracting single-scale features often refined by stacked layered attentions. This paper presents novel deep learning-based Multiscale Feature Enhancement via Dual Attention Network specifically designed for retinal OCT images. Our approach leverages EfficientNetB7 backbone to extract multiscale from images, ensuring comprehensive representation global local structures. To further refine feature extraction, we propose Pyramidal mechanism that integrates Multi-Head Self-Attention (MHSA) with Dense Atrous Spatial Pyramid Pooling (DASPP), effectively capturing long-range dependencies contextual information at multiple scales. Additionally, Efficient Channel (ECA) Refinement modules are introduced enhance channel-wise spatial representations, enabling precise localization abnormalities. A ablation study confirms progressive impact integrated blocks attention mechanisms overall performance. findings underscore potential advanced processing, highlighting effectiveness network. Extensive experiments two benchmark datasets demonstrate superiority proposed network over existing state-of-the-art methods.

Language: Английский

Incorporation of Histogram Intersection and Semantic Information into Non-Negative Local Laplacian Sparse Coding for Image Classification DOI Creative Commons
Ying Shi, Yuan Wan, Xinjian Wang

et al.

Mathematics, Journal Year: 2025, Volume and Issue: 13(2), P. 219 - 219

Published: Jan. 10, 2025

Traditional sparse coding has proven to be an effective method for image feature representation in recent years, yielding promising results classification. However, it faces several challenges, such as sensitivity variations, code instability, and inadequate distance measures. Additionally, classification often operate independently, potentially resulting the loss of semantic relationships. To address these issues, a new is proposed, called Histogram intersection Semantic information-based Non-negativity Local Laplacian Sparse Coding (HS-NLLSC) This integrates Locality into (NLLSC) optimisation, enhancing stability ensuring that similar features are encoded codewords. In addition, histogram introduced redefine between vectors codebooks, effectively preserving their similarity. By comprehensively considering both processes classification, more information retained, thereby leading representation. Finally, multi-class linear Support Vector Machine (SVM) employed Experimental on four standard three maritime datasets demonstrate superior performance compared previous six algorithms. Specifically, accuracy our approach improved by 5% 19% methods. research provides valuable insights various stakeholders selecting most suitable specific circumstances.

Language: Английский

Citations

1
HDL-ACO hybrid deep learning and ant colony optimization for ocular optical coherence tomography image classification DOI Creative Commons
Shivani Agarwal, Anand Kumar Dohare,

Pranshu Saxena

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 18, 2025

Language: Английский

Citations

0
Enhancing Ocular Health Precision: Cataract Detection Using Fundus Images and ResNet-50 DOI
Irshad Khan,

Wajahat Akbar,

Abdullah Abdullah

et al.

IECE transactions on intelligent systematics., Journal Year: 2024, Volume and Issue: 1(3), P. 145 - 160

Published: Oct. 29, 2024

Cataracts are a leading cause of blindness in Pakistan, contributing to more than 54% cases due poor living condition, nutritional deficiencies, and limited healthcare access. Early detection is critical avoid invasive treatments,but current diagnostic approaches often identify cataracts at advanced stages. This paper presents an advanced,automated cataract system using deep learning specifically the ResNet-50 architecture, address this gap. The model processes fundus retinal images curated from diverse datasets, classified by ophthalmologic experts through rigorous three-stage process. By leveraging model, categorized into normal,moderate,and severe, achieving accuracy 97.56% on full images. Notably, performs well even partial with 70% visibility, maintaining 95.23%, thus minimizing need for extensive restoration. dataset was augmented include 17,500 images,ensuring robust training. model's ability detect high precision varying visibility(70% ,80%,85% beyond) demonstrate its flexibility reliability, consistently above 95.50%. research offers non-invasive, efficient solution particularly suited remote areas, addressing limitations late-stage diagnoses. It represent significant advancement has potential revolutionize global identification early, accurate intervention.

Language: Английский

Citations

2
Multiscale attention-over-attention network for retinal disease recognition in OCT radiology images DOI Creative Commons
Abdulmajeed M. Alenezi,

Daniyah A. Aloqalaa,

Sushil Kumar Singh

et al.

Frontiers in Medicine, Journal Year: 2024, Volume and Issue: 11

Published: Nov. 8, 2024

Retinal disease recognition using Optical Coherence Tomography (OCT) images plays a pivotal role in the early diagnosis and treatment of conditions. However, previous attempts relied on extracting single-scale features often refined by stacked layered attentions. This paper presents novel deep learning-based Multiscale Feature Enhancement via Dual Attention Network specifically designed for retinal OCT images. Our approach leverages EfficientNetB7 backbone to extract multiscale from images, ensuring comprehensive representation global local structures. To further refine feature extraction, we propose Pyramidal mechanism that integrates Multi-Head Self-Attention (MHSA) with Dense Atrous Spatial Pyramid Pooling (DASPP), effectively capturing long-range dependencies contextual information at multiple scales. Additionally, Efficient Channel (ECA) Refinement modules are introduced enhance channel-wise spatial representations, enabling precise localization abnormalities. A ablation study confirms progressive impact integrated blocks attention mechanisms overall performance. findings underscore potential advanced processing, highlighting effectiveness network. Extensive experiments two benchmark datasets demonstrate superiority proposed network over existing state-of-the-art methods.

Language: Английский

Citations

0