Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit
The detection of Diabetic Retinopathy (DR) is an emergent research topic in recent decades, where DR is a primary cause of vision loss in humans. The existing techniques have limitations such as neuron death issues, vanishing gradient, and output offset. To overcome these issues, this paper propose...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IIUM Press, International Islamic University Malaysia
2025-01-01
|
| Series: | International Islamic University Malaysia Engineering Journal |
| Subjects: | |
| Online Access: | https://journals.iium.edu.my/ejournal/index.php/iiumej/article/view/3194 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841549931014258688 |
|---|---|
| author | Pravin Balaso Chopade Prabhakar N. Kota Bhagvat D. Jadhav Pravin Marotrao Ghate Shriram Sadashiv Kulkarni |
| author_facet | Pravin Balaso Chopade Prabhakar N. Kota Bhagvat D. Jadhav Pravin Marotrao Ghate Shriram Sadashiv Kulkarni |
| author_sort | Pravin Balaso Chopade |
| collection | DOAJ |
| description |
The detection of Diabetic Retinopathy (DR) is an emergent research topic in recent decades, where DR is a primary cause of vision loss in humans. The existing techniques have limitations such as neuron death issues, vanishing gradient, and output offset. To overcome these issues, this paper proposes a Deep Learning (DL)-based technique for early and accurate DR detection. The Coordinate Attention Module (CAM) based Convolutional Neural Network (CNN) with Leaky Rectified Linear Unit (LReLU) is proposed for early and accurate detection of DR. The MESSIDOR dataset is preprocessed through the median filter to eliminate noise, and Contrast-Limited Adaptive Histogram Equalization (CLAHE) is utilized to increase the contrast level in an input image. The preprocessed images are given to Mayfly Optimization Algorithm-based Region Growing (MOARG) for image segmentation. Then, the features are extracted using ResNet50 and SqueezeNet, which extract deep learning features. The extracted features are given to CAM-based CNN with LReLU to detect DR, which overcomes the dead issues of neurons and minimizes the probability of inactive neurons. The proposed model achieves better results on the MESSIDOR datasets on the metrics of accuracy, precision, recall, specificity, f1-score, and Area Under Curve (AUC) values of about 99.72%, 99.46%, 99.25%, 99.61%, 99.37% and 99.14%, correspondingly, proving to be superior to the existing method, Capsule Network and Hybrid Adaptive DL based DR (HADL-DR).
ABSTRAK: Pengesanan Retinopati Diabetik (DR) merupakan topik penyelidikan yang semakin mendapat perhatian dalam dekad-dekad kebelakangan ini, di mana DR merupakan punca utama kehilangan penglihatan pada manusia. Teknik sedia ada mempunyai beberapa kekangan seperti isu kematian neuron, vanishing gradient, dan output offset. Untuk mengatasi isu-isu ini, kertas ini mencadangkan teknik berasaskan Pembelajaran Mendalam (DL) untuk pengesanan awal dan tepat bagi DR. Modul Coordinate Attention Module (CAM) berasaskan Convolutional Neural Network (CNN) dengan Leaky Rectified Linear Unit (LReLU) dicadangkan untuk pengesanan awal dan tepat bagi DR. Dataset MESSIDOR diproses melalui penapis median yang digunakan untuk menghapuskan hingar, dan Contrast-Limited Adaptive Histogram Equalization (CLAHE) digunakan untuk meningkatkan tahap kontras pada imej input. Imej yang telah diproses diberikan kepada Algoritma Pengoptimuman Mayfly berasaskan Region Growing (MOARG) untuk segmentasi imej. Kemudian, ciri-ciri diekstrak menggunakan ResNet50 dan SqueezeNet yang mengekstrak ciri-ciri pembelajaran mendalam. Ciri-ciri yang diekstrak ini diberikan kepada CNN berasaskan CAM dengan LReLU untuk pengesanan DR, yang mengatasi isu kematian neuron dan meminimumkan kebarangkalian neuron tidak aktif. Model yang dicadangkan mencapai keputusan yang lebih baik pada dataset MESSIDOR berdasarkan metrik ketepatan, ketepatan, panggilan semula, kekhususan, skor f1, dan nilai Kawasan di Bawah Lengkung (AUC) iaitu sekitar 99.72%, 99.46%, 99.25%, 99.61%, 99.37% dan 99.14%, masing-masing, membuktikan keunggulannya berbanding kaedah sedia ada, Capsule Network dan Hybrid Adaptive DL berasaskan DR (HADL-DR).
|
| format | Article |
| id | doaj-art-6cc94d8e36764f6dba5971182b82bd1a |
| institution | Kabale University |
| issn | 1511-788X 2289-7860 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IIUM Press, International Islamic University Malaysia |
| record_format | Article |
| series | International Islamic University Malaysia Engineering Journal |
| spelling | doaj-art-6cc94d8e36764f6dba5971182b82bd1a2025-01-10T12:40:45ZengIIUM Press, International Islamic University MalaysiaInternational Islamic University Malaysia Engineering Journal1511-788X2289-78602025-01-0126110.31436/iiumej.v26i1.3194Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear UnitPravin Balaso Chopade0Prabhakar N. Kota1https://orcid.org/0000-0002-7537-8433Bhagvat D. Jadhav2https://orcid.org/0000-0002-1393-6823Pravin Marotrao Ghate3Shriram Sadashiv Kulkarni4M.E.S. College of EngineeringM.E.S. College of EngineeringJSPM's Rajarshi Shahu College of EngineeringJSPM's Rajarshi Shahu College of EngineeringSinhgad Academy of Engineering The detection of Diabetic Retinopathy (DR) is an emergent research topic in recent decades, where DR is a primary cause of vision loss in humans. The existing techniques have limitations such as neuron death issues, vanishing gradient, and output offset. To overcome these issues, this paper proposes a Deep Learning (DL)-based technique for early and accurate DR detection. The Coordinate Attention Module (CAM) based Convolutional Neural Network (CNN) with Leaky Rectified Linear Unit (LReLU) is proposed for early and accurate detection of DR. The MESSIDOR dataset is preprocessed through the median filter to eliminate noise, and Contrast-Limited Adaptive Histogram Equalization (CLAHE) is utilized to increase the contrast level in an input image. The preprocessed images are given to Mayfly Optimization Algorithm-based Region Growing (MOARG) for image segmentation. Then, the features are extracted using ResNet50 and SqueezeNet, which extract deep learning features. The extracted features are given to CAM-based CNN with LReLU to detect DR, which overcomes the dead issues of neurons and minimizes the probability of inactive neurons. The proposed model achieves better results on the MESSIDOR datasets on the metrics of accuracy, precision, recall, specificity, f1-score, and Area Under Curve (AUC) values of about 99.72%, 99.46%, 99.25%, 99.61%, 99.37% and 99.14%, correspondingly, proving to be superior to the existing method, Capsule Network and Hybrid Adaptive DL based DR (HADL-DR). ABSTRAK: Pengesanan Retinopati Diabetik (DR) merupakan topik penyelidikan yang semakin mendapat perhatian dalam dekad-dekad kebelakangan ini, di mana DR merupakan punca utama kehilangan penglihatan pada manusia. Teknik sedia ada mempunyai beberapa kekangan seperti isu kematian neuron, vanishing gradient, dan output offset. Untuk mengatasi isu-isu ini, kertas ini mencadangkan teknik berasaskan Pembelajaran Mendalam (DL) untuk pengesanan awal dan tepat bagi DR. Modul Coordinate Attention Module (CAM) berasaskan Convolutional Neural Network (CNN) dengan Leaky Rectified Linear Unit (LReLU) dicadangkan untuk pengesanan awal dan tepat bagi DR. Dataset MESSIDOR diproses melalui penapis median yang digunakan untuk menghapuskan hingar, dan Contrast-Limited Adaptive Histogram Equalization (CLAHE) digunakan untuk meningkatkan tahap kontras pada imej input. Imej yang telah diproses diberikan kepada Algoritma Pengoptimuman Mayfly berasaskan Region Growing (MOARG) untuk segmentasi imej. Kemudian, ciri-ciri diekstrak menggunakan ResNet50 dan SqueezeNet yang mengekstrak ciri-ciri pembelajaran mendalam. Ciri-ciri yang diekstrak ini diberikan kepada CNN berasaskan CAM dengan LReLU untuk pengesanan DR, yang mengatasi isu kematian neuron dan meminimumkan kebarangkalian neuron tidak aktif. Model yang dicadangkan mencapai keputusan yang lebih baik pada dataset MESSIDOR berdasarkan metrik ketepatan, ketepatan, panggilan semula, kekhususan, skor f1, dan nilai Kawasan di Bawah Lengkung (AUC) iaitu sekitar 99.72%, 99.46%, 99.25%, 99.61%, 99.37% dan 99.14%, masing-masing, membuktikan keunggulannya berbanding kaedah sedia ada, Capsule Network dan Hybrid Adaptive DL berasaskan DR (HADL-DR). https://journals.iium.edu.my/ejournal/index.php/iiumej/article/view/3194Area Under CurveConvolutional Neural NetworkCoordinate Attention ModuleLeakyLeaky Rectified Linear Unit, Rectified Linear Unit,Mayfly Optimization Algorithm |
| spellingShingle | Pravin Balaso Chopade Prabhakar N. Kota Bhagvat D. Jadhav Pravin Marotrao Ghate Shriram Sadashiv Kulkarni Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit International Islamic University Malaysia Engineering Journal Area Under Curve Convolutional Neural Network Coordinate Attention Module LeakyLeaky Rectified Linear Unit, Rectified Linear Unit, Mayfly Optimization Algorithm |
| title | Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit |
| title_full | Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit |
| title_fullStr | Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit |
| title_full_unstemmed | Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit |
| title_short | Retinopathy Disease Detection and Classification Using a Coordinate Attention Module-Based Convolutional Neural Network with Leaky Rectified Linear Unit |
| title_sort | retinopathy disease detection and classification using a coordinate attention module based convolutional neural network with leaky rectified linear unit |
| topic | Area Under Curve Convolutional Neural Network Coordinate Attention Module LeakyLeaky Rectified Linear Unit, Rectified Linear Unit, Mayfly Optimization Algorithm |
| url | https://journals.iium.edu.my/ejournal/index.php/iiumej/article/view/3194 |
| work_keys_str_mv | AT pravinbalasochopade retinopathydiseasedetectionandclassificationusingacoordinateattentionmodulebasedconvolutionalneuralnetworkwithleakyrectifiedlinearunit AT prabhakarnkota retinopathydiseasedetectionandclassificationusingacoordinateattentionmodulebasedconvolutionalneuralnetworkwithleakyrectifiedlinearunit AT bhagvatdjadhav retinopathydiseasedetectionandclassificationusingacoordinateattentionmodulebasedconvolutionalneuralnetworkwithleakyrectifiedlinearunit AT pravinmarotraoghate retinopathydiseasedetectionandclassificationusingacoordinateattentionmodulebasedconvolutionalneuralnetworkwithleakyrectifiedlinearunit AT shriramsadashivkulkarni retinopathydiseasedetectionandclassificationusingacoordinateattentionmodulebasedconvolutionalneuralnetworkwithleakyrectifiedlinearunit |