GabROP: Gabor Wavelets-Based CAD for Retinopathy of Prematurity Diagnosis via Convolutional Neural Networks

Abstract

One of the most serious and dangerous ocular problems in premature infants is retinopathy of prematurity (ROP), a proliferative vascular disease. Ophthalmologists can use automatic computer-assisted diagnostic (CAD) tools to help them make a safe, accurate, and low-cost diagnosis of ROP. All previous CAD tools for ROP diagnosis use the original fundus images. Unfortunately, learning the discriminative representation from ROP-related fundus images is difficult. Textural analysis techniques, such as Gabor wavelets (GW), can demonstrate significant texture information that can help artificial intelligence (AI) based models to improve diagnostic accuracy. In this paper, an effective and automated CAD tool, namely GabROP, based on GW and multiple deep learning (DL) models is proposed. Initially, GabROP analyzes fundus images using GW and generates several sets of GW images. Next, these sets of images are used to train three convolutional neural networks (CNNs) models independently. Additionally, the actual fundus pictures are used to build these networks. Using the discrete wavelet transform (DWT), texture features retrieved from every CNN trained with various sets of GW images are combined to create a textural-spectral-temporal demonstration. Afterward, for each CNN, these features are concatenated with spatial deep features obtained from the original fundus images. Finally, the previous concatenated features of all three CNN are incorporated using the discrete cosine transform (DCT) to lessen the size of features caused by the fusion process. The outcomes of GabROP show that it is accurate and efficient for ophthalmologists. Additionally, the effectiveness of GabROP is compared to recently developed ROP diagnostic techniques. Due to GabROP's superior performance compared to competing tools, ophthalmologists may be able to identify ROP more reliably and precisely, which could result in a reduction in diagnostic effort and examination time.

Keywords: Gabor wavelets (GW); artificial intelligence (AI); computer assisted diagnosis (CAD); deep learning; eye disease; retinopathy of prematurity (ROP).

Figure 1

Examples of the dataset’s images, (a) diseased and (b), not diseased.

Figure 2

The stages of the proposed GabROP CAD tool.

Figure 3

Samples of the generated GW images for both classes of the ROP dataset, (a) Diseased, (b) Not Diseased.

Figure 4

Diagnostic accuracy of the three classifiers trained with the features extracted from ResNet-50 learned using the individual GW images compared the fused DWT features.

Figure 5

Diagnostic accuracy of the three classifiers trained with the features extracted from DarkNet-53 learned using the individual GW images compared to the fused DWT features.

Figure 6

Diagnostic accuracy of the three classifiers trained with the features extracted from MobileNet learned using the individual GW images compared to the fused DWT features.

Figure 7

Diagnostic accuracy of the three classifiers trained with spatial features extracted from ResNet-50 learned using the original fundus images compared the fused DWT features obtained by CNNs learned with GW images and the combination of the two.

Figure 8

Diagnostic accuracy of the three classifiers trained with spatial features extracted from DarkNet-53 learned using the original fundus images compared the fused DWT features obtained by CNNs learned with GW images and the combination of the two.

Figure 9

Diagnostic accuracy of the three classifiers trained with spatial features extracted from MobileNet learned using the original fundus images compared the fused DWT features obtained by CNNs learned with GW images and the combination of the two.

Figure 10

Diagnostic accuracy of the three classifiers trained with integrated features of the third fusion stage (fusing of the second fusion stage features of the three CNNs using DCT) versus the number of DCT features.

Figure 11

ROC curve and the AUC of the SVM classifier trained with integrated features of the third fusion stage (fusing of the second fusion stage features of the three CNNs using DCT (2000 features)).

Figure 12

Comparison among the highest accuracy attained in each fusion stage of GabROP.