ABSTRACT
Background: Accurate classification techniques are essential for the early diagnosis and treatment of patients with diabetic retinopathy (DR). However, the limited amount of annotated DR data poses a challenge for existing deep-learning models. This article proposes a difficulty-aware and task-augmentation method based on meta-learning (DaTa-ML) model for few-shot DR classification with fundus images. Methods: The difficulty-aware (Da) method operates by dynamically modifying the cross-entropy loss function applied to learning tasks. This methodology has the ability to intelligently down-weight simpler tasks, while simultaneously prioritizing more challenging tasks. These adjustments occur automatically and aim to optimize the learning process. Additionally, the task-augmentation (Ta) method is used to enhance the meta-training process by augmenting the number of tasks through image rotation and improving the feature-extraction capability. To implement the expansion of the meta-training tasks, various task instances can be sampled during the meta-training stage. Ultimately, the proposed Ta method was introduced to optimize the initialization parameters and enhance the meta-generalization performance of the model. The DaTa-ML model showed promising results by effectively addressing the challenges associated with few-shot DR classification. Results: The Asia Pacific Tele-Ophthalmology Society (APTOS) 2019 blindness detection data set was used to evaluate the DaTa-ML model. The results showed that with only 1% of the training data (5-way, 20-shot) and a single update step (training time reduced by 90%), the DaTa-ML model had an accuracy rate of 89.6% on the test data, which is a 1.7% improvement over the transfer-learning method [i.e., residual neural network (ResNet)50 pre-trained on ImageNet], and a 16.8% improvement over scratch-built models (i.e., ResNet50 without pre-trained weights), despite having fewer trainable parameters (the parameters used by the DaTa-ML model are only 0.47% of the ResNet50 parameters). Conclusions: The DaTa-ML model provides a more efficient DR classification solution with little annotated data and has significant advantages over state-of-the-art methods. Thus, it could be used to guide and assist ophthalmologists to determine the severity of DR.
ABSTRACT
PURPOSE: Diabetic retinopathy (DR) is one of the most serious complications of diabetes, which is a kind of fundus lesion with specific changes. Early diagnosis of DR can effectively reduce the visual damage caused by DR. Due to the variety and different morphology of DR lesions, automatic classification of fundus images in mass screening can greatly save clinicians' diagnosis time. To alleviate these problems, in this paper, we propose a novel framework-graph attentional convolutional neural network (GACNN). METHODS AND MATERIALS: The network consists of convolutional neural network (CNN) and graph convolutional network (GCN). The global and spatial features of fundus images are extracted by using CNN and GCN, and attention mechanism is introduced to enhance the adaptability of GCN to topology map. We adopt semi-supervised method for classification, which greatly improves the generalization ability of the network. RESULTS: In order to verify the effectiveness of the network, we conducted comparative experiments and ablation experiments. We use confusion matrix, precision, recall, kappa score, and accuracy as evaluation indexes. With the increase of the labeling rates, the classification accuracy is higher. Particularly, when the labeling rate is set to 100%, the classification accuracy of GACNN reaches 93.35%. Compared with DenseNet121, the accuracy rate is improved by 6.24%. CONCLUSIONS: Semi-supervised classification based on attention mechanism can effectively improve the classification performance of the model, and attain preferable results in classification indexes such as accuracy and recall. GACNN provides a feasible classification scheme for fundus images, which effectively reduces the screening human resources.
