EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy
Present architecture of convolution neural network for diabetic retinopathy (DR-Net) is based on normal convolution (NC). It incurs high computational cost as NC uses a multiplicative weight that measures a combined correlation in both cross-channel and spatial dimension of layer's inputs. This...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Published: |
Institute of Electrical and Electronics Engineers Inc.
2022
|
| Subjects: |
| _version_ | 1796867657684746240 |
|---|---|
| author | Aujih, Ahmad Bukhari Shapiai, Mohd. Ibrahim Meriaudeau, Fabrice Tang, Tong Boon |
| author_facet | Aujih, Ahmad Bukhari Shapiai, Mohd. Ibrahim Meriaudeau, Fabrice Tang, Tong Boon |
| author_sort | Aujih, Ahmad Bukhari |
| collection | ePrints |
| description | Present architecture of convolution neural network for diabetic retinopathy (DR-Net) is based on normal convolution (NC). It incurs high computational cost as NC uses a multiplicative weight that measures a combined correlation in both cross-channel and spatial dimension of layer's inputs. This might cause the overall DR-Net architecture to be over-parameterised and computationally inefficient. This paper proposes EDR-Net-a new end-to-end, DR-Net architecture with depth-wise separable convolution module. The EDR-Net architecture was trained with DRKaggle-train dataset (35,126 images), and tested on two datasets, i.e. DRKaggle-test (53,576 images) and Messidor-2 (1,748 images). Results showed that the proposed EDR-Net achieved predictive performance comparable with current state-of-the-arts in detecting referable diabetic retinopathy (rDR) from fundus images and outperformed other light weight architectures, with at least two times less computation cost. This makes it more amenable for mobile device based computer-assisted rDR screening applications. |
| first_indexed | 2024-03-05T21:30:34Z |
| format | Article |
| id | utm.eprints-104439 |
| institution | Universiti Teknologi Malaysia - ePrints |
| last_indexed | 2024-03-05T21:30:34Z |
| publishDate | 2022 |
| publisher | Institute of Electrical and Electronics Engineers Inc. |
| record_format | dspace |
| spelling | utm.eprints-1044392024-02-04T10:04:26Z http://eprints.utm.my/104439/ EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy Aujih, Ahmad Bukhari Shapiai, Mohd. Ibrahim Meriaudeau, Fabrice Tang, Tong Boon T Technology (General) Present architecture of convolution neural network for diabetic retinopathy (DR-Net) is based on normal convolution (NC). It incurs high computational cost as NC uses a multiplicative weight that measures a combined correlation in both cross-channel and spatial dimension of layer's inputs. This might cause the overall DR-Net architecture to be over-parameterised and computationally inefficient. This paper proposes EDR-Net-a new end-to-end, DR-Net architecture with depth-wise separable convolution module. The EDR-Net architecture was trained with DRKaggle-train dataset (35,126 images), and tested on two datasets, i.e. DRKaggle-test (53,576 images) and Messidor-2 (1,748 images). Results showed that the proposed EDR-Net achieved predictive performance comparable with current state-of-the-arts in detecting referable diabetic retinopathy (rDR) from fundus images and outperformed other light weight architectures, with at least two times less computation cost. This makes it more amenable for mobile device based computer-assisted rDR screening applications. Institute of Electrical and Electronics Engineers Inc. 2022 Article PeerReviewed Aujih, Ahmad Bukhari and Shapiai, Mohd. Ibrahim and Meriaudeau, Fabrice and Tang, Tong Boon (2022) EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy. IEEE Transactions on Biomedical Circuits and Systems, 16 (3). pp. 467-478. ISSN 1932-4545 http://dx.doi.org/10.1109/TBCAS.2022.3182907 DOI : 10.1109/TBCAS.2022.3182907 |
| spellingShingle | T Technology (General) Aujih, Ahmad Bukhari Shapiai, Mohd. Ibrahim Meriaudeau, Fabrice Tang, Tong Boon EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title | EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title_full | EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title_fullStr | EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title_full_unstemmed | EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title_short | EDR-Net: Lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| title_sort | edr net lightweight deep neural network architecture for detecting referable diabetic retinopathy |
| topic | T Technology (General) |
| work_keys_str_mv | AT aujihahmadbukhari edrnetlightweightdeepneuralnetworkarchitecturefordetectingreferablediabeticretinopathy AT shapiaimohdibrahim edrnetlightweightdeepneuralnetworkarchitecturefordetectingreferablediabeticretinopathy AT meriaudeaufabrice edrnetlightweightdeepneuralnetworkarchitecturefordetectingreferablediabeticretinopathy AT tangtongboon edrnetlightweightdeepneuralnetworkarchitecturefordetectingreferablediabeticretinopathy |