تشخیص و درجه‌بندی خودکار رتینوپاتی دیابتی با استفاده از شبکه کانولوشنی سلسله‌مراتبی و مکانیزم توجه دوگانه

مقالات آماده انتشار

نوع مقاله : مقاله پژوهشی

نویسندگان

1 دانشجوی دکتری دانشکده مهندسی کامپیوتر، دانشگاه صنعتی شاهرود، شاهرود، ایران

2 دانشیار دانشکده مهندسی کامپیوتر، دانشگاه صنعتی شاهرود، شاهرود، ایران

چکیده

رتینوپاتی دیابتی یکی از شایع‌ترین عوارض چشمی دیابت و عامل اصلی نابینایی بزرگسالان است. تشخیص زودهنگام آن به دلیل ابعاد کوچک ضایعات اولیه (مانند میکروآنوریسم‌ها) و شباهت با بافت سالم شبکیه، چالشی اساسی است. روش‌های متداول یادگیری عمیق در ایجاد تعادل میان استخراج جزئیات بافتی و درک معنایی کلی محدودیت دارند. در این مقاله برای غلبه بر این محدودیت، شبکه‌ای کانولوشنی با ساختار دوشاخه‌ای پیشنهادشده که از محو شدن جزئیات ریز ضایعات در لایه‌های عمیق جلوگیری می‌کند. این معماری از شبکه ResNet-50 پیش‌آموخته به‌دلیل ساختار باقیمانده و توانایی استخراج ویژگی‌های چندسطحی بهره برده و اطلاعات را در دو مسیر مدیریت می‌کند. مسیر اول ویژگی‌های سطح میانی (جزئیات محلی) و مسیر دوم ویژگی‌های معنایی عمیق را پردازش می‌کند. ماژول توجه فضایی و کانالی تعبیه‌شده در شبکه، با مدل‌سازی وابستگی‌ها، به‌صورت انطباقی نواحی ضایعه را تقویت و نویز پس‌زمینه را سرکوب می‌کند. همچنین از تابع ضرر ترکیبی (آنتروپی متقاطع و فوکال) برای مدیریت عدم‌تعادل داده‌ها و تمرکز بر نمونه‌های دشوار استفاده‌شده است. ارزیابی روی دادگان APTOS 2019 صحت 23/97% در تشخیص باینری و 06/83% در درجه‌بندی ۵ کلاسه را نشان داد. این نتایج در حالت چندکلاسه، بیانگر بهبود ۲ درصدی صحت و ۴ درصدی امتیاز F1 نسبت به بهترین روش‌های پیشرفته اخیر است. نتایج تجربی کارایی بالای مدل پیشنهادی را اثبات می‌کند.

کلیدواژه‌ها

موضوعات

عنوان مقاله [English]

Automatic Detection and Grading of Diabetic Retinopathy Using a Hierarchical Convolutional Network and Dual Attention Mechanism

نویسندگان [English]

  • Mustapha Mejman Salem Alghanami 1
  • Mansoor Fateh 2
1 PHD Candidate, Faculty of Computer Engineering, Shahrood University of Technology, Shahrood, Iran
2 Associate Professor, Faculty of Computer Engineering, Shahrood University of Technology, Shahrood, Iran
چکیده [English]

Diabetic Retinopathy is one of the most common ocular complications of diabetes and a leading cause of blindness among adults. Early detection of this disease poses a significant challenge in medical image processing due to the minute size of early-stage lesions, such as microaneurysms, and their visual similarity to healthy retinal tissue. Conventional deep learning methods often face limitations in balancing the extraction of fine-grained textural details with the comprehension of global semantic information. In this paper, we propose a Hierarchical Convolutional Neural Network with a Dual Attention Mechanism for the detection and grading of Diabetic Retinopathy. The proposed architecture utilizes a pre-trained ResNet-50 backbone for feature extraction and manages information flow through two distinct pathways. The first pathway focuses on extracting local details by integrating mid-level features, while the second pathway processes deep semantic features. To enhance feature discriminability, both pathways are equipped with a Hierarchical Spatial and Channel Attention Module that adaptively highlights lesion-related regions and suppresses background noise. Furthermore, to address the class imbalance problem, a hybrid loss function combining Cross-Entropy and Focal Loss is employed. Evaluation on the standard APTOS 2019 dataset demonstrates that the proposed method achieves an accuracy of %97.23 in binary classification (Healthy vs. DR) and %83.06 in multi-class severity grading (5-class). Experimental results and ablation studies confirm the efficacy of the dual-branch architecture and attention modules compared to recent state-of-the-art methods

کلیدواژه‌ها [English]

  • Diabetic Retinopathy
  • Deep Learning
  • ResNet
  • Attention Mechanism
  • Medical Image Analysis

مراجع

[1]        N. Jagadisha, A. P. Prabhu, S. Abhay, C. Nayak, and C. S. Honavarker, "Retinaguard: An optimal cnn solution for diabetic retinopathy," in 2024 International Conference on Knowledge Engineering and Communication Systems (ICKECS), vol. 1. IEEE, 2024, pp. 1-7.
[2]        Kumar, L. Nelson, and D. Arumugam, "Advanced cnn-based approach for early diabetic retinopathy detection using retinal images," in 2024 International Conference on Artificial Intelligence and Emerging Technology (Global AI Summit). IEEE, 2024, pp. 399-404.
[3]        Mushtaq, Gazala, and Farheen Siddiqui. "Detection of diabetic retinopathy using deep learning methodology." IOP conference series: materials science and engineering. Vol. 1070. No. 1. IOP Publishing, 2021.
[4]        S. Zhu, C. Xiong, Q. Zhong, and Y. Yao, "Diabetic retinopathy classification with deep learning via fundus images: A short survey," IEEE Access, vol. 12, pp. 20540-20558, 2024.
[5]        Lin, Chun-Ling, and Kun-Chi Wu. "Development of revised ResNet-50 for diabetic retinopathy detection." BMC bioinformatics 24.1 (2023): 157.
[6]        Mutawa, S. Alnajdi, and S. Sruthi, "Transfer learning for diabetic retinopathy detection: A study of dataset combination and model performance," Applied Sciences, vol. 13, no. 9, p. 5685, 2023.
[7]        M. Youldash, et al., "Early detection and classification of diabetic retinopathy: A deep learning approach," AI, vol. 5, no. 4, 2024.
[8]        M. V. Kumar, et al., "Diabetic retinopathy classification using quantum-assisted deep learning," in Evolution in Computational Intelligence: FICTA 2024, Springer Nature, 2025.
[9]        Wang, Hao, et al. "Many-MobileNet: Multi-model Augmentation for Robust Retinal Disease Classification." MICCAI Challenge on Ultra-Widefield Fundus Imaging for Diabetic Retinopathy. Cham: Springer Nature Switzerland, 2024. 144-154.
[10]    R. Bala, A. Sharma, and N. Goel, "Comparative analysis of diabetic retinopathy classification approaches using machine learning and deep learning techniques," Archives of Computational Methods in Engineering, vol. 31, no. 2, pp. 919-955, 2024.
[11]       Zhang, Yiye, et al. "Development and validation of a machine learning algorithm for predicting the risk of postpartum depression among pregnant women." Journal of affective disorders 279 (2021): 1-8.
[12]    G. Tăbăcaru, et al., "A robust machine learning model for diabetic retinopathy classification," Journal of Imaging, vol. 10, no. 1, p. 8, 2023.
[13]    S. Lakhera and A. Garg, "Diabetic retinopathy classification using psosvm based deep learning model," in 2023 Seventh International Conference on Image Information Processing (ICIIP). IEEE, 2023.
[14]    P. K. Ramanaiah, et al., "Diabetic retinopathy classification using various machine learning techniques," TPM-Testing, Psychometrics, Methodology in Applied Psychology, vol. 32, no. S5, pp. 1175-1183, 2025.
[15]    M. Youldash, et al., "Early detection and classification of diabetic retinopathy: A deep learning approach," AI, vol. 5, no. 4, 2024.
[16]    A. Mutawa, S. Alnajdi, and S. Sruthi, "Transfer learning for diabetic retinopathy detection: A study of dataset combination and model performance," Applied Sciences, vol. 13, no. 9, p. 5685, 2023.
[17]    A. Jabbar, et al., "A lesion-based diabetic retinopathy detection through hybrid deep learning model," IEEE Access, vol. 12, pp. 40019-40036, 2024.
[18]    M. Santos, et al., "Predicting diabetic retinopathy stage using siamese convolutional neural network," Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, vol. 12, no. 1, p. 2297017, 2024.
[19]    M. Touati, et al., "Drect: Enhancing diabetic retinopathy classification with a compact convolutional transformer," Big Data and Cognitive Computing, vol. 9, no. 1, p. 9, 2025.
[20]    H. Xu, et al., "A hybrid neural network approach for classifying diabetic retinopathy subtypes," Frontiers in Medicine, vol. 10, p. 1293019, 2024.
[21]    Kakade, Gaurang, and Chaitanya Kakade. "Optimal Detection of Diabetic Retinopathy Severity Levels Using Attention-Based CNN and Vision Transformers (ViT)." 2024 International Conference on Modeling, Simulation & Intelligent Computing (MoSICom). IEEE, 2024.
[22]    Z. Liu, et al., "Identification of diabetic retinopathy lesions in fundus images by integrating cnn and vision mamba models," PLoS One, vol. 20, no. 1, p. e0318264
[23] Asia Pacific Tele-Ophthalmology Society, “Aptos 2019 blindness detection,” https://www.kaggle.com/c/ aptos2019-blindness-detection, 2019, accessed: 2024.
[24] P. Porwal, S. Pachade, R. Kamble, M. Kokare, G. Deshmukh, V. Sahasrabuddhe, F. Meriaudeau, Indian diabetic retinopathy image dataset (IDRiD): a database for diabetic retinopathy screening research, Data, 3(3) (2018) 25.
[25] Khan, Zubair, et al. "Diabetic retinopathy detection using VGG-NIN a deep learning architecture." IEEE Access 9 (2021): 61408-61416.
[26] Sikder, Niloy, et al. "Severity classification of diabetic retinopathy using an ensemble learning algorithm through analyzing retinal images." Symmetry 13.4 (2021): 670.
[27] Folorunsho, O., et al. "Explainable ensemble deep learning model for predicting diabetic retinopathy based on APTOS 2019 eye pack dataset." LAUTECH Journal of Engineering and Technology 19.1 (2025): 1-14.
[28] P. Chen, “Jetson nano v1: Resnet34 implementation,” Kaggle Notebook, 2024. [Online]. Available: https://www.kaggle.com/code/peiyuchen412/jetson-nano-v1.
[29] V. Palan, “Final project f3: Efficientnet experiments,”Kaggle Notebook, 2024. [Online]. Available: https://www.kaggle.com/code/vatsalpalan99/final-proj-f3.
[30]    A. Ikram, A. I. Mudassir, and Abdullah, “Diabeticretinopathy classification with densenet201,” Kaggle Notebook, 2024. [Online]. Available: https://www.kaggle.com/code/liveinthemoment/notebook5f0e0b81de.
[31]    D. Malhotra, “Qcnn-dr: Quantum convolutional neural network,” Kaggle Notebook, 2024. [Online]. Available: https://www.kaggle.com/code/dakshm/qcnn-dr
[32]    Shi, Bojia, et al. "GoogLeNet-based Diabetic-retinopathy-detection." 2022 14th International Conference on Advanced Computational Intelligence (ICACI). IEEE, 2022.
[33]    Mohanty, Cheena, et al. "Using deep learning architectures for detection and classification of diabetic retinopathy." Sensors 23.12 (2023): 5726.
[34]    Gangwar, Akhilesh Kumar, and Vadlamani Ravi. "Diabetic retinopathy detection using transfer learning and deep learning." Evolution in Computational Intelligence: Frontiers in Intelligent Computing: Theory and Applications (FICTA 2020), Volume 1. Singapore: Springer Singapore, 2020. 679-689.
[35]    Y. Yang, A. Panagopoulou, S. Zhou, D. Jin, C. Callison-Burch, M. Yatskar, Language in a bottle: Language model guided concept bottlenecks for interpretable image classification, in, 2023, pp. 19187--19197.
[36]    M. Yuksekgonul, M. Wang, J. Zou, Post-hoc concept bottleneck models, arXiv preprint arXiv:2205.15480, (2022).
[37]    Y. Gao, D. Gu, M. Zhou, D. Metaxas, Aligning human knowledge with visual concepts towards explainable medical image classification, in, 2024, pp. 46-56.
پژوهش های نظری و کاربردی هوش ماشینی

مقالات آماده انتشار، پذیرفته شده
انتشار آنلاین از تاریخ 31 مرداد 1405

فایل ها

هم رسانی

ارجاع به این مقاله

آمار