Retinal Image Quality Classification Using Fine-Tuned CNN

Sun, Jing; Wan, Cheng; Cheng, Jun; Yu, Fengli; Liu, Jiang

doi:10.1007/978-3-319-67561-9_14

Cited by 10 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Blind IQA has become more generalizable with the development of machine learning and does not require additional information beyond the original data [14]. As such, it is commonly used for fundus images classification [15][16][17][18][19][20][21] and has proven to be effective in eliminating low-quality images. Few studies have investigated the use of blind IQA methods based on OCT until 2019.…”

Section: Introductionmentioning

confidence: 99%

Deep learning for quality assessment of retinal OCT images

Wang

Deng

et al. 2019

Biomed. Opt. Express

View full text Add to dashboard Cite

Optical coherence tomography (OCT) is a promising high-speed, non-invasive imaging modality providing high-resolution retinal scans. However, a variety of external factors such as light occlusion and patient movement can seriously degrade OCT image quality, which complicates manual retinopathy detection and computer-aided diagnosis. As such, this study first presents an OCT image quality assessment (OCT-IQA) system, capable of automatic classification based on signal completeness, location, and effectiveness. Four CNN architectures (VGG-16, Inception-V3, ResNet-18, and ResNet-50) from the ImageNet classification task were used to train the proposed OCT-IQA system via transfer learning. The ResNet-50 with the best performance was then integrated into the final OCT-IQA network. The usefulness of this approach was evaluated using retinopathy detection results. A retinopathy classification network was first trained by fine-tuning Inception-V3 model. The model was then applied to two test datasets, created randomly from the original dataset, one of which was screened by the OCT-IQA system and only included high quality images while the other was mixed by high and low quality images. Results showed that retinopathy detection accuracy and area under curve (AUC) were 3.75% and 1.56% higher, respectively, for the filtered data (compared with the unfiltered data). These experimental results demonstrate the effectiveness of the proposed OCT-IQA system and suggest that deep learning could be applied to the design of computer-aided systems (CADSs) for automatic retinopathy detection.

show abstract

Section: Introductionmentioning

confidence: 99%

Deep learning for quality assessment of retinal OCT images

Wang

Deng

et al. 2019

Biomed. Opt. Express

View full text Add to dashboard Cite

show abstract

“…Sun et al [25] evaluated the various CNN (VGG-16, ResNet-50, AlexNet and GoogLeNet) performances in the classification of retinal images. This model evaluated the two major factors, namely pre-processing and data augmentation.…”

Section: Related Workmentioning

confidence: 99%

An Image Quality Selection and Effective Denoising on Retinal Images Using Hybrid Approaches

Saju,

2023

IJRITCC

View full text Add to dashboard Cite

Retinal image analysis has remained an essential topic of research in the last decades. Several algorithms and techniques have been developed for the analysis of retinal images. Most of these techniques use benchmark retinal image datasets to evaluate performance without first exploring the quality of the retinal image. Hence, the performance metrics evaluated by these approaches are uncertain. In this paper, the quality of the images is selected by utilizing the hybrid naturalness image quality evaluator and the perception-based image quality evaluator (hybrid NIQE-PIQE) approach. Here, the raw input image quality score is evaluated using the Hybrid NIQE-PIQE approach. Based on the quality score value, the deep learning convolutional neural network (DCNN) categorizes the images into low quality, medium quality and high quality images. Then the selected quality images are again pre-processed to remove the noise present in the images. The individual green channel (G-channel) is extracted from the selected quality RGB images for noise filtering. Moreover, hybrid modified histogram equalization and homomorphic filtering (Hybrid G-MHE-HF) are utilized for enhanced noise filtering. The implementation of proposed scheme is implemented on MATLAB 2021a. The performance of the implemented method is compared with the other approaches to the accuracy, sensitivity, specificity, precision and F-score on DRIMDB and DRIVE datasets. The proposed scheme’s accuracy is 0.9774, sensitivity is 0.9562, precision is 0.99, specificity is 0.99, and F-measure is 0.9776 on the DRIMDB dataset, respectively.

show abstract

“…On a larger dataset (9653 ungradable retinal images and 11347 gradable images), they also evaluated the possibility of using a hybrid method combining saliency maps and CNNs [44]. Finally, [45] compare the performance of fine-tuning four CNN architectures -AlexNet [35], GoogLeNet [46], VGG-16 [47] and ResNet-50 -on a 3000-image subset of the Kaggle database. These preliminary studies report that large networks are hard to train, and must deal with overfitting issues, due to the huge amount of parameters.…”

Section: Related Workmentioning

confidence: 99%