Multivariate Regression-Based Convolutional Neural Network Model for Fundus Image Quality Assessment

Abstract

Objectively assessing the perceptual quality of an ocular fundus image is essential for the reliable diagnosis of various ocular diseases. A fair amount of work has been done in this field to date. However, the generalizability of the current work is limited, as the existing quality models were developed and evaluated with data-sets built with limited subjective inputs. This paper aims at addressing this limitation with the following two contributions. First, a new fundus image quality assessment (FIQuA) data-set is presented, containing 1500 fundus images with three classes of quality: Good, Fair, and Poor. Also, for each image, subjective scores (in the range [0-10]) were collected for six quality parameters, including structural and generic properties of the fundus images. Second, a new multivariate regression based convolutional neural network (CNN) model is proposed to predict the fundus image quality. The proposed model consists of two individually trained blocks. The first block consists of four pre-trained models, trained against the subjective scores for the six quality parameters, and aims at deriving the optimized features for classification. Next, the optimized features from each of the four models are ensembled together and transferred to the second block for final classification. The proposed model achieves a strong correlation with the subjective scores, with the values 0.941, 0.954, 0.853, and 0.401 obtained for SROCC, LCC, KCC, and RMSE respectively. Its classification accuracy is 95.66% over the FIQuA data-set, and 98.96% and 88.43% respectively over the two publicly available data-sets DRIMDB and EyeQ. INDEX TERMS Fundus image quality assessment, diabetic retinopathy, multivariate regression, convolutional neural network.

Colorectal Cancer Segmentation using Atrous Convolution and Residual Enhanced UNet

“…The power of CNNs was first exhibited in the ImageNet challenge [6], and ever since then, CNNs have revolutionized the field of computer vision and have produced far better results than conventional techniques on numerous tasks. The same can be said in the case of medical imaging [19], particularly segmentation. The inherent property of CNNs to automatically find crucial and task-relevant structures in images account for their widespread use.…”

A Deep Retinal Image Quality Assessment Network with Salient Structure Priors

“…Motivated by the success of CNN in computer vision tasks, CNN has been introduced to Retinal-IQA. Several methods for Retinal-IQA are either training a shallow classification CNN network from scratch [36] or fine-tuning existing classification networks, such as AlexNet [12], ResNet [14] and Xception [37], from pre-trained models [3,38,39]. To make use of the complementary of different colour spaces, Fu et al [4] propose a multiple colour-space fusion network, in which multiple parallel CNN branches are used to learn features from different colour-spaces.…”

“…We compare our SalStructIQA with six image quality assessment methods: BRISQUE [19], NBIQA [20] and TS-CNN [27], HVS-based method [8], DenseNet121-MCF [4] and Multivariate-Regression CNN (MR-CNN) [39]. The first three methods are designed for Natural-IQA, among which BRISQUE [19] and NBIQA [20] are hand-crafted feature based and TS-CNN [27] is deep feature based.…”

“…In the implementation of TS-CNN [27], the regression layer with one-dimensional output is replaced with a classification layer with three-dimensional outputs and the loss is replaced by multi-class crossentropy loss. The last three methods are specifically designed for Retinal-IQA, among which HVS-based method [8] is hand-crafted feature based and DenseNet121-MCF [4] and MR-CNN [39] are deep feature based. The results of HVS-based method [8] and DenseNet121-MCF [4] are from [4].…”

Small sample color fundus image quality assessment based on gcforest

“…Saha [29] constructed a database and developed a deep learning framework to assess image quality in the context of diabetic retinopathy. Raj [24] proposed a new multivariate regression based convolutional neural network model to predict the fundus image quality. For the same reason, they constructed a new priveta database include 1500 fundus images that is classified three class, namely good, fair, poor.…”