Detection of Diabetic Eye Disease from Retinal Images Using a Deep Learning Based CenterNet Model

Abstract

Diabetic retinopathy (DR) is an eye disease that alters the blood vessels of a person suffering from diabetes. Diabetic macular edema (DME) occurs when DR affects the macula, which causes fluid accumulation in the macula. Efficient screening systems require experts to manually analyze images to recognize diseases. However, due to the challenging nature of the screening method and lack of trained human resources, devising effective screening-oriented treatment is an expensive task. Automated systems are trying to cope with these challenges; however, these methods do not generalize well to multiple diseases and real-world scenarios. To solve the aforementioned issues, we propose a new method comprising two main steps. The first involves dataset preparation and feature extraction and the other relates to improving a custom deep learning based CenterNet model trained for eye disease classification. Initially, we generate annotations for suspected samples to locate the precise region of interest, while the other part of the proposed solution trains the Center Net model over annotated images. Specifically, we use DenseNet-100 as a feature extraction method on which the one-stage detector, CenterNet, is employed to localize and classify the disease lesions. We evaluated our method over challenging datasets, namely, APTOS-2019 and IDRiD, and attained average accuracy of 97.93% and 98.10%, respectively. We also performed cross-dataset validation with benchmark EYEPACS and Diaretdb1 datasets. Both qualitative and quantitative results demonstrate that our proposed approach outperforms state-of-the-art methods due to more effective localization power of CenterNet, as it can easily recognize small lesions and deal with over-fitted training data. Our proposed framework is proficient in correctly locating and classifying disease lesions. In comparison to existing DR and DME classification approaches, our method can extract representative key points from low-intensity and noisy images and accurately classify them. Hence our approach can play an important role in automated detection and recognition of DR and DME lesions.

Making Artificial Intelligence Lemonade Out of Data Lemons

“…While simply duplicating videos leads to algorithm performance failure due to overfitting, image augmentations such as rotation, inversions, contrast changes, and others can result in a more robust algorithm than would be possible otherwise with limited data. 38 Not only is the difference in anatomic appearance between the A4C and SX imaging windows inherently a challenge, but also simply using data from a different type of ultrasound machine can markedly decrease performance for algorithms. 39 Both were factors at play in the current research.…”

“…Video rotations and flips allowed the researchers to convert a 191 video dataset into 3820 videos used for algorithm training. While simply duplicating videos leads to algorithm performance failure due to overfitting, image augmentations such as rotation, inversions, contrast changes, and others can result in a more robust algorithm than would be possible otherwise with limited data 38 …”

An Augmented Artificial Intelligence Approach for Chronic Diseases Prediction

“…This study showed that pre-processing of data prior to model building enhances prediction accuracy ( 35 ). Machine learning methods are also used for the detection of some other medical diseases detection ( 36 – 38 ), text mining ( 39 – 42 ) and network security ( 43 – 47 ). Another analysis is conducted to predict the risks associated with diabetes.…”

DFTSA-Net: Deep Feature Transfer-Based Stacked Autoencoder Network for DME Diagnosis

“…The DCNN model achieved an accuracy of 91.2%, specificity of 94.4%, and sensitivity of 87.5%. A custom deep learning-based CenterNet model was developed by Nazir et al [36] for various DR lesions' classification. Their method involved dataset preparation and feature extraction using DenseNet-100.…”

“…In horizontal reflection, all image columns are flipped around the vertical axis, as shown in Figure 4. Training a deep learning model with an augmented dataset has been shown to improve the performance of the model, reduce overfitting, and increase the model's ability to generalize [36]. In order to balance the augmented dataset, the same number of images for each class was used, providing a total number of 888 images.…”