Myopia has become a major public health problem affecting the eye health of our citizens, especially teenagers. Fundus images can be obtained non-invasively and can be used to monitor and follow up on the progress in high myopia. However, with the development of artificial intelligence, it is still difficult to establish a computer-aided diagnosis model for high myopia with young children as research objects, mainly because 1) it is very difficult to collect labeled fundus images, and there is no large amount of such data that can be freely accessed; 2) hard samples which have clinical significance for population screening and diagnosis are rare and indistinguishable. To solve these problems, we propose a high myopia detection model on fundus images based on curriculum learning. We design a dual-curriculum generation module, which aims to use the expert model to endow the curriculum with new training indicators so that the student model can gradually and robustly identify hard samples. Compared with the baseline, our framework significantly improves the convergence speed of the training process and achieves the best performance during testing. Experimental results on a high myopia fundus images dataset show that our framework provides efficient and accurate detection and outperforms other methods.

Cataract is a chronic eye disease that causes irreversible vision loss. Automatic cataract detection can help people prevent visual impairment and decrease the possibility of blindness. To date, many studies utilize deep learning methods to grade cataract severity on fundus images. However, they mainly focus on the classification performance and ignore the model interpretability, which may lead to a semantic gap between networks and users. In this paper, we present a deep learning network to improve the model interpretability, which consists three main modules: deep feature extraction, visual saliency module and semantic description module. Visual and semantic interpretation jointly employed to provide cataract-grade oriented interpretation for the overall model. Experimental results on real clinical data set show that our method improves the interpretability for cataract grading while ensuring the high classification performance.

Hydronephrosis may lead to many potential diseases, and the diagnosis of hydronephrosis is time-consuming and laborious. To assist physicians in hydronephrosis diagnosis and treatment planning, an accurate and automatic kidney segmentation method is highly required in clinical practice. In recent years, deep convolutional neural networks such as Unet plays a key role in the field of image segmentation, but Unet itself cannot adjust the receptive field actively, which may result in poor attention to the characteristics of the segmented target. We propose an encoder-decoder network with weighted skip connections and the idea of hierarchical equal resolution that can manually control the receptive field. We evaluated our method by comparing it with various classical networks using a dataset of 1850 annotated images. The MPA of the model is 94.12 and the MIoU is 89.49, which outperformed other classical networks we compared to.

Hydronephrosis is a common renal disease in children which can lead to a series of complications, and ultrasonography is a basic examination usually performed on suspected hydronephrosis patients. If we can use deep learning approaches to judge and grade the disease in the ultrasonic examination stage, we can save a lot of manpower, medical resources, money, and help the suffered patients. For the semantic segmentation of kidney ultrasound image, we designed an Attention-based Pyramid Scene Parsing Network (A-PSPNet), the core of which is the basic feature extraction network combining Convolutional Block Attention Module (CBAM) and pyramid analysis module. Experiments were carried out on a hydronephrosis dataset containing 1850 annotated ultrasound images, including the arrangement of attention units, statistical computing power, and comparison of the effectiveness between the benchmark and our proposed method. Our constructed model achieved better segmentation performance than benchmarks with only little extra overhead, which validated the lightweight and effectiveness of the model.

Cataract is one of the most serious eye diseases leading to blindness. Early detection and treatment can reduce the rate of blindness in cataract patients. However, the professional knowledge of ophthalmologists is necessary for the clinical cataract detection. Therefore, the potential costs may make it difficult for the widespread use of cataract detection to prevent blindness. Artificial intelligence assisted diagnosis based on medical images has attracted more and more attention of researchers. Many studies have focused on the use of pre-defined feature sets for cataract classification, but the predefined feature sets may be incomplete or redundant. On account of the aforementioned issues, some studies have proposed deep learning methods to automatically extract image features, but all based on global features and none has analyzed the layer-by-layer transformation process of the middle-tier features. This paper uses convolutional neural networks (CNN) to learn useful features directly from input data, and deconvolution network method is employed to investigate how CNN characterizes cataract layer-by-layer. We found that compared to the global feature set, the detail vascular information, which is lost after multi-layer convolution calculation also plays an important role in cataract grading task. And this finding fits with the morphological definition of fundus image. Through the finding, we gained insights into the design of hybrid global-local feature representation model to improve the recognition performance of automatic cataract grading.

Cataract is defined as a lenticular opacity presenting usually with poor visual acuity. It is considered the most common cause of blindness. Early diagnosis and treatment can reduce the suffering of patients and prevent visual impairment from turning into blindness. Recently, cataract diagnosis applying pattern recognition is in a rising period. For retinal fundus images, the task is usually cataract classification. However, it needs complex manual processing, which demands dexterous people and time taking exertion. Besides, it faces the challenge of effective interpretability and dependability. In this paper, we develop a deep-learning algorithm to intuitively identify cataract attributes to solve these limitations. Our model, is a 18(50)-layer convolutional neural network that inputs retinal image in G channel and outputs the prediction with heatmap. The heatmap localizes the areas where most indicative of different levels of cataract. Furthermore, we extend the training strategy for the corresponding task, which aims at improving the performance of the network. Comparing with other methods in cataract classification, we succeeded to achieve state of the art accuracy of proposed method on detection and grading task. Most importantly, our model provides a compelling reason via localizing the areas revealing cataract in the image.