The causal relationships among depression, hypertriglyceridemia, and cardiomyopathy remain undefined. Cardiomyopathy encompasses both ischemic heart disease (IHD) and dilated cardiomyopathy (DCM). Inverse variance weighted random-effect, inverse variance weighted fixed-effect, maximum likelihood estimation, MR-Egger, weighted median, and penalized weighted median were employed to evaluate the causal relationship between exposure and outcome. Stepwise testing and multivariable Mendelian randomization analyses were employed in the mediation analysis. The onset of depression precipitates an increase in triglyceride levels, which in turn escalates the risk of developing DCM. The onset of depression not only directly escalates the risk of developing IHD but also indirectly amplifies this risk by elevating triglyceride levels. Exacerbation of depression was observed to lead to an elevation in triglyceride levels (odds ratio [OR]: 1.093, 95% confidence interval [CI]: 1.041-1.149, P < .001). There was no direct causal link established between depression and DCM (P = .392). The incidence of depression was associated with an increased risk of developing IHD (OR: 1.019, 95% CI: 1.011-1.026, P < .001). Elevated triglyceride levels were found to augment the risk of both IHD (OR: 1.019, 95% CI: 1.015-1.023, P < .001) and DCM (OR: 1.240, 95% CI: 1.047-1.469, P = .013). In the causal pathway between depression and DCM, triglycerides were found to mediate 100% of the effect. In the relationship between depression and IHD, triglycerides accounted for a mediation proportion of 8.4%.

Objective To investigate methods for constructing a high-quality instructional dataset for traditional Chinese medicine (TCM) mental disorders and to validate its efficacy. Methods We proposed the Fine-Med-Mental-T&P methodology for constructing high-quality instruction datasets in TCM mental disorders. This approach integrates theoretical knowledge and practical case studies through a dual-track strategy. (i) Theoretical track: textbooks and guidelines on TCM mental disorders were manually segmented. Initial responses were generated using DeepSeek-V3, followed by refinement by the Qwen3-32B model to align the expression with human preferences. A screening algorithm was then applied to select 16 000 high-quality instruction pairs. (ii) Practical track: starting from over 600 real clinical case seeds, diagnostic and therapeutic instruction pairs were generated using DeepSeek-V3 and subsequently screened through manual evaluation, resulting in 4 000 high-quality practice-oriented instruction pairs. The integration of both tracks yielded the Med-Mental-Instruct-T&P dataset, comprising a total of 20 000 instruction pairs. To validate the dataset’s effectiveness, three experimental evaluations (both manual and automated) were conducted: (i) comparative studies to compare the performance of models fine-tuned on different datasets; (ii) benchmarking to compare against mainstream TCM-specific large language models (LLMs); (iii) data ablation study to investigate the relationship between data volume and model performance. Results Experimental results demonstrate the superior performance of T&P-model fine-tuned on the Med-Mental-Instruct-T&P dataset. In the comparative study, the T&P-model significantly outperformed the baseline models trained solely on self-generated or purely human-curated baseline data. This superiority was evident in both automated metrics (ROUGE-L > 0.55) and expert manual evaluations (scoring above 7/10 across accuracy). In benchmark comparisons, the T&P-model also excelled against existing mainstream TCM LLMs (e.g., HuatuoGPT and ZuoyiGPT). It showed particularly strong capabilities in handling diverse clinical presentations, including challenging disorders such as insomnia and coma, showcasing its robustness and versatility. Data ablation studies showed that T&P-model performance had an overall upward trend with minor fluctuations when training data increased from 10% to 50%; beyond 50%, performance improvement slowed significantly, with metrics plateauing and approaching a saturation point. Conclusion This study has successfully constructed the specialized Med-Mental-Instruct-T&P instruction dataset for TCM mental disorders proposed the systematic Fine-Med-Mental-T&P methodology for its development, effectively addressing the critical challenge of high-quality, domain-specific data scarcity in TCM, and providing essential data support for developing intelligent TCM diagnostic and therapeutic systems.

Generating therapeutic descriptions for Traditional Chinese Medicine (TCM) herbs from images is challenging due to subtle inter-class visual variations, domain-specific knowledge complexity, and the lack of large-scale benchmark datasets. Therapeutic descriptions are essential in TCM as they provide information on the functions, properties, and clinical applications of herbs, supporting identification, education, and clinical decision-making. To the best of our knowledge, this is the first study to formulate TCM therapeutic description generation as an image captioning task. We propose a novel image captioning framework, UVA-Cap, which introduces an improved attention mechanism called Upgraded Visual Attention (UVA). UVA enriches conventional attention by fusing global visual context with localized feature selection, resulting in a more comprehensive and context-sensitive visual representation that significantly improves the quality, coherence, and semantic consistency of generated captions. In addition to introducing the novel model UVA-Cap, the framework employs a ResNet-50-based CNN to extract both global and local visual features and evaluates multiple captioning architectures, including Single LSTM variants, Dual-LSTM models, and the Up-Down model. To support this task, we construct the TCM-TheraCap dataset with 130,381 images of 80 TCM herb species, each paired with detailed therapeutic annotations. Experimental results show that the proposed approach generates accurate and semantically meaningful therapeutic descriptions, establishing a benchmark for future research in TCM herb understanding.

Microaneurysms (MAs) are pivotal biomarkers of early diabetic retinopathy (DR), and their accurate segmentation in fundus images is indispensable for timely screening and intervention. To alleviate the annotation burden inherent to fully supervised approaches, we propose a semi-supervised segmentation framework termed CBAMCPS that leverages cross pseudo-supervision. The architecture employs a dual-branch PFFNet backbone where: (i) one branch generates high-confidence pseudolabels via multi-scale feature fusion, while (ii) the other embeds CBAM attention modules at encoder - decoder keypoints to capture spatial - channel saliencies. A Combo Loss is further introduced to counteract class imbalance. Trained on the IDRiD dataset with only 81 annotated and 1 439 unannotated images ($\approx 1/8$ of full supervision), CBAMCPS attains 71.3% mIoU and 77.6% mDice, substantially outperforming the fully supervised baseline ($65.0 \% \text{mIoU}$). These results substantiate the efficacy of unlabeled data exploitation and the indispensability of each proposed component. CBAM-CPS establishes an efficient and scalable semi-supervised paradigm for MA segmentation under extremely low-annotation regimes, paving the way for largescale DR screening.

Background The prescriptions of traditional chinese medicine (TCM) have made a great contribution to the treatment of disease and the maintenance of good health. Current research on prescription recommendations mainly focuses on the correlation between symptoms and herbs. However, the semantic information inherent in both symptoms and herbs has received limited attention. Furthermore, most datasets in the field of TCM suffer from limited data volumes, which can adversely impact model training. Methods To tackle these challenges, we present a prescription recommendation framework called PRDAGE, which is based on data augmentation and multi-graph embedding. We started by collecting medical records and creating a dataset of 3,052 classic medical cases, where we normalized the symptoms and herbs. Additionally, we developed a multi-layer embedding method for symptoms and herbs, using Sentence Bert (SBert) and graph convolutional networks. The aim of this multi-layer embedding method is to capture and represent the semantic information of symptoms and herbs, as well as the complex relationships between them. Additionally, a median-based random data augmentation method was introduced to enrich the medical case data, effectively enhancing the model’s accuracy. Results The model was evaluated against baseline models on an unenhanced dataset (Dataset-B), and the results showed that the proposed PRDAGE framework exhibited superior overall performance. Compared to the second-best model, PRDAGE achieved improvements in accuracy and recall rates of 1.69% and 3.80%, respectively, on the Top@10 metric. Ablation experiments further revealed that both the data augmentation and multi-layer embedding modules contributed to the improved model performance. Conclusion In conclusion, the experimental results suggest that PRDAGE is an effective prescription recommendation framework. The multi-layer embedding approach effectively represents the semantic information of symptoms and the complex relationships between symptoms and herbs. Additionally, the use of median-based data augmentation has a positive impact on the overall performance and generalization ability of the model.

AIM: To develop a traditional Chinese medicine (TCM) knowledge graph (KG) for diabetic retinopathy (DR) diagnosis and treatment by integrating literature and medical records, thereby enhancing TCM knowledge accessibility and providing innovative approaches for TCM inheritance and DR management. METHODS: First, a KG framework was established with a schema-layer design. Second, high-quality literature and electronic medical records served as data sources. Named entity recognition was performed using the ALBERT-BiLSTM-CRF model, and semantic relationships were curated by domain experts. Third, knowledge fusion was mainly achieved through an alias library. Subsequently, the data layer was mapped to the schema layer to refine the KG, and knowledge was stored in Neo4j. Finally, exploratory work on intelligent question answering was conducted based on the constructed KG. RESULTS: In Neo4j, a KG for TCM diagnosis and treatment was constructed, incorporating 6 types of labels, 5 types of relationships, 5 types of attributes, 822 nodes, and 1,318 relationship instances. This systematic KG supports logical reasoning and intelligent question answering. The question answering model achieved a precision of 95%, a recall of 95%, and a weighted F1-score of 95%. CONCLUSION: This study proposes a semi-automatic knowledge-mapping scheme to balance integration efficiency and accuracy. Clinical data-driven entity and relationship construction enables digital dialectical reasoning. Exploratory applications show the KG's potential in intelligent question answering, providing new insights for TCM health management.

Objective To determine the correlation between traditional Chinese medicine (TCM) inspection of spirit classification and the severity grade of depression based on facial features, offering insights for intelligent intergrated TCM and western medicine diagnosis of depression. Methods Using the Audio-Visual Emotion Challenge and Workshop (AVEC 2014) public dataset on depression, which conclude 150 interview videos, the samples were classified according to the TCM inspection of spirit classification: Deshen (得神, presence of spirit), Shaoshen (少神, insufficiency of spirit), and Shenluan (神乱, confusion of spirit). Meanwhile, based on Beck Depression Inventory-II (BDI-II) score for the severity grade of depression, the samples were divided into minimal (0 – 13, Q1), mild (14 – 19, Q2), moderate (20 – 28, Q3), and severe (29 – 63, Q4). Sixty-eight landmarks were extracted with a ResNet-50 network, and the feature extracion mode was stadardized. Random forest and support vectior machine (SVM) classifiers were used to predict TCM inspection of spirit classification and the severity grade of depression, respectively. A Chi-square test and Apriori association rule mining were then applied to quantify and explore the relationships. Results The analysis revealed a statistically significant and moderately strong association between TCM spirit classification and the severity grade of depression, as confirmed by a Chi-square test (χ2 = 14.04, P = 0.029) with a Cramer’s V effect size of 0.243. Further exploration using association rule mining identified the most compelling rule: “moderate depression (Q3) → Shenluan”. This rule demonstrated a support level of 5%, indicating this specific co-occurrence was present in 5% of the cohort. Crucially, it achieved a high Confidence of 86%, meaning that among patients diagnosed with Q3, 86% exhibited the Shenluan pattern according to TCM assessment. The substantial Lift of 2.37 signifies that the observed likelihood of Shenluan manifesting in Q3 patients is 2.37 times higher than would be expected by chance if these states were independent—compelling evidence of a highly non-random association. Consequently, Shenluan emerges as a distinct and core TCM diagnostic manifestation strongly linked to Q3, forming a clinically significant phenotype within this patient subgroup. Conclusion Automated facial analysis can serve as a common lens for TCM and western psychological assessments align in the diagnosis of depression. The inspection of spirit decline trajectory parallels worsening depression, supporting early screening and stratified intervention, and providing a reference for the intelligent assistance of integrated TCM and western medicine in the diagnosis of depression.

>In the wave of digital and intelligent applications, artificial intelligence (AI) is transforming the development trajectories of industries across the globe. Traditional Chinese medicine (TCM), as a cultural treasure of the Chinese nation, carries thousands of years of wisdom and practical experience. However, in the context of the rapid advancements in modern medicine and technology,

Some observational studies have highlighted a significant association between cholecystitis and factors leading to sudden death; however, the specific relationship between the 2 has not been fully elucidated. The primary objective of this study was to elucidate the causal interplay between cholecystitis and augmented risk of sudden cardiac death. We used large-scale genetic summary data from genome-wide association study, genetic summary statistics were sourced from 3 eminent repositories: the UK Biobank (N = 463,010), the FinnGen consortium (N = 215,027), and the European Bioinformatics Institute (N = 471,251). By employing 2-sample Mendelian randomization (MR) to decipher the causal interplay between cholecystitis and sudden death etiologies, a meta-analytical approach was employed to amalgamate the findings derived from these disparate data sources. The primary MR methodologies used included inverse variance weighting with random effects, inverse variance weighting with fixed effects, maximum likelihood, MR-Egger, and weighted median. Subsequently, we performed heterogeneity testing, polyvalency examination, and sensitivity analysis to bolster the robustness of causal relationship assessments. Meta-analysis and amalgamating variegated data sources revealed a statistically significant inverse correlation between cholecystitis and ventricular arrhythmias (odds ratio, 0.896; 95% confidence interval: 0.826–0.971; P = .008). Similarly, an inverse association was observed between cholecystitis and aortic aneurysm (odds ratio, 0.899; 95% confidence interval: 0.851–0.951, P < .001). This study substantiates the absence of a direct causal link between cholecystitis and cerebrovascular accidents (P = .771), pulmonary embolism (P = .071), and acute myocardial infarction (P = .388). A direct causal correlation existed between cholecystitis and sudden death associated with ventricular arrhythmias and aortic aneurysms. The onset of cholecystitis may mitigate the risk of sudden death due to ventricular arrhythmias and aortic aneurysms.

Objective:To mine the medication patterns of ancient prescriptions for diabetic retinopathy(DR)from databases of traditional Chinese medicine(TCM)ancient books,and provide evidence for clinical practice and scientific research of TCM treatment for DR. Methods:The traditional library retrieval and modern data retrieval technology were combined to collect the ancient prescriptions in these databases,including the library of Hunan University of Chinese Medicine,Chinese Medical Dictionary,Duxiu,and Chaoxing Digital Library.And the TCM inheritance auxiliary platform(V3.0)was used for data mining,mainly including drug frequency analysis,medicinal property and meridian tropism analysis,efficacy analysis,correlation analysis,complex network analysis,and cluster analysis. Results:A total of 271 ancient prescriptions for the treatment of DR were collected,involving 296 drugs.The total medication frequency was 2,727.Most of them were cold and sweet drugs.The meridians primarily targeted were the liver,kidney,and spleen.The main effects of drugs were supplementing deficiency,clearing heat,releasing the exterior,inducing urination to drain dampness,pacifying liver and extinguishing wind,and circulating blood and transforming stasis.Saposhnikovia divaricata was the most frequently Chinese herbal medicine for DR in TCM ancient books.Saposhnikovia divaricata and ligusticum wallichii,saposhnikovia divaricata and notopterygium root,angelica sinensis and ligusticum wallichii were common herbal pairs.Saposhnikovia divaricata,ginseng,plantain seed,angelica sinensis,prepared rehmannia root and cassia seed constituted the core formula with the highest frequency. Conclusion:The core prescriptions for treating DR are mainly crafted from Dihuang pill,Ruiren powder,Siwu decoction,and Zhujing pill.Saposhnikovia divaricata is an important meridian-guiding medicine to open Xuanfu for DR.In clinical practice,the prescriptions should be modified according to the evolution of pathogenesis.

Age-related macular degeneration (AMD) and diabetic macular edema (DME) are significant causes of blindness worldwide. The prevalence of these diseases is steadily increasing due to population aging. Therefore, early diagnosis and prevention are crucial for effective treatment. Classification of Macular Degeneration OCT Images is a widely used method for assessing retinal lesions. However, there are two main challenges in OCT image classification: incomplete image feature extraction and lack of prominence in important positional features. To address these challenges, we proposed a deep learning neural network model called MSA-Net, which incorporates our proposed multi-scale architecture and spatial attention mechanism. Our multi-scale architecture is based on depthwise separable convolution, which ensures comprehensive feature extraction from multiple scales while minimizing the growth of model parameters. The spatial attention mechanism is aim to highlight the important positional features in the images, which emphasizes the representation of macular region features in OCT images. We test MSA-NET on the NEH dataset and the UCSD dataset, performing three-class (CNV, DURSEN, and NORMAL) and four-class (CNV, DURSEN, DME, and NORMAL) classification tasks. On the NEH dataset, the accuracy, sensitivity, and specificity are 98.1%, 97.9%, and 98.0%, respectively. After fine-tuning on the UCSD dataset, the accuracy, sensitivity, and specificity are 96.7%, 96.7%, and 98.9%, respectively. Experimental results demonstrate the excellent classification performance and generalization ability of our model compared to previous models and recent well-known OCT classification models, establishing it as a highly competitive intelligence classification approach in the field of macular degeneration.

Diabetic retinopathy (DR) is a chronic condition that can lead to significant vision loss and even blindness. Existing deep networks for hard exudate segmentation in fundus images face two primary challenges: (1) The receptive field of traditional convolution operations is limited, resulting in poor hard exudate extraction performance; (2) Due to the irregular distribution and different sizes of fine exudates, it is easy to lose information about tiny exudates during the feature extraction process. To address these challenges, we propose DBASNet, a novel lesion segmentation model. In order to solve the problem of insufficient segmentation caused by the limitations of the receptive field, we propose a new multi-scale attention feature extraction (MAT) module. Combined with the dual encoder structure, the features extracted by MAT and EfficientNet in the dual branches are fused to effectively expand the perceptual field and avoid information loss. We also propose an attentional skip connection (AS) module in the decoder to filter and retain channel and spatial information, enrich skip connections and carry feature information of tiny lesions. Experiments on publicly available datasets IDRiD and E-Ophtha-EX demonstrate effectiveness of our method. DBASNet achieves 79.48, 80.35, 79.81, and 66.64% of recall, precision, Dice, and IOU metrics on IDRiD and 52.73, 60.33, 56.16, and 39.82% on E-Ophtha-EX, respectively. DBASNet outperforms some state-of-the-art approaches. The quantitative and qualitative findings unequivocally establish the pre-eminence of DBASNet in the field of lesion segmentation relevant to diabetic retinopathy.

Abstract Retinal microaneurysm (MA) is a definite earliest clinical sigh of diabetic retinopathy (DR). Its automatic segmentation is key to realizing intelligent screening for early DR, which can significantly reduce the risk of visual impairment in patients. However, the minute scale and subtle contrast of MAs against the background pose challenges for segmentation. This paper focuses on automatic MA segmentation in fundus images. A novel pyramid feature fusion network (PFFNet) that progressively develops and fuses rich contextual information by integrating two pyramid modules is proposed. Multiple global pyramid scene parsing (GPSP) modules are introduced between the encoder and decoder to provide diverse global contextual information for the decoder through reconstructing skip connections. Additionally, a spatial scale‐aware pyramid (SSAP) module is introduced to dynamically fuse multi‐scale contextual information. This rich contextual information will help to identify MAs from low‐contrast background. Furthermore, to mitigate issue related to category imbalance, a combo loss function is introduced. Finally, to validate the effectiveness of the proposed method, experiments are conducted on two publicly available datasets, IDRiD and DDR, and PFFNet is compared with several state‐of‐the‐art models. The experimental results demonstrate the superiority of our PFFNet in the MA segmentation task.

Eye diagnosis is a method for inspecting systemic diseases and syndromes by observing the eyes.With the development of intelligent diagnosis in traditional Chinese medicine(TCM),artificial intelligence(AI)can improve the accuracy and efficiency of eye diagnosis.However,the research on intelligent eye diagnosis still faces many challenges,including the lack of standardized and precisely labeled data,multi-modal information analysis,and artificial in-telligence models for syndrome differentiation.The widespread application of AI models in medicine provides new insights and opportunities for the research of eye diagnosis intelli-gence.This study elaborates on the three key technologies of AI models in the intelligent ap-plication of TCM eye diagnosis,and explores the implications for the research of eye diagno-sis intelligence.First,a database concerning eye diagnosis was established based on self-su-pervised learning so as to solve the issues related to the lack of standardized and precisely la-beled data.Next,the cross-modal understanding and generation of deep neural network models to address the problem of lacking multi-modal information analysis.Last,the build-ing of data-driven models for eye diagnosis to tackle the issue of the absence of syndrome dif-ferentiation models.In summary,research on intelligent eye diagnosis has great potential to be applied the surge of AI model applications.

In order to take full advantage of traditional Chinese medicine (TCM) and western medicine, combined with machine learning technology, to study the risk factors and better risk prediction model of diabetic retinopathy (DR), and provide basis for the screening and treatment of it. Through a retrospective study of DR cases in the real world, the electronic medical records of patients who met screening criteria were collected. Moreover, Recursive Feature Elimination with Cross-Validation (RFECV) was used for feature selection. Then, the prediction model was built based on Gradient Boosting Machine (GBM) and it was compared with 4 other popular machine learning techniques, including Logistic Regression (LR), K-Nearest Neighbors (KNN), Random Forest, and Support Vector Machine (SVM). The models were evaluated with accuracy, precision, recall, F1 score, and area under the curve (AUC) value as indicators. In addition, grid search was used to optimize the model. To explain the results of the model more intuitively, the Shapley Additive exPlanation (SHAP) method was used. A total of 9034 type 2 diabetes mellitus (T2DM) patients meeting the screening criteria were included in this study, including 1118 patients with DR. 19 features were selected using RFECV in the model construction. We constructed 5 commonly used models, including GBM, LR, KNN, Random Forest, and SVM. By comparing model performance, GBM has the highest accuracy (0.85) and AUC value (0.934), which is the best prediction model. We also carried out hyperparameter optimization of grid search for this model, and the model accuracy reached 0.88, and the AUC value increased to 0.958. Through SHAP analysis, it was found that TCM syndrome types, albumin, low density lipoprotein, triglyceride, total protein, glycosylated hemoglobin were closely related to the increased risk of DR. It can be concluded that TCM syndrome type is the risk factor of DR. The GBM classifier based on grid search optimization, with relevant risk factors of TCM and western medicine as variables, can better predict the risk of DR.

Objective: A plethora of studies have unequivocally established the profound significance of harmonious familial relationships on the psychological well-being of the elderly. In this study, we elucidate the intergenerational relationships, probing the association between frequent interactions or encounters with their children and the incidence of depression in old age. Methodology: We employed a retrospective cross-sectional study design, sourcing our data from the 2018 wave of the China Health and Retirement Longitudinal Study (CHARLS). To identify cases of depression, we utilized the 10-item Center for Epidemiologic Studies Depression Scale (CES-D). Employing a five-fold cross-validation methodology, we endeavored to fashion five distinct machine learning models. Subsequently, we crafted learning curves to facilitate the refinement of hyperparameters, assessing model classification performance through metrics such as accuracy and the Area Under the Receiver Operating Characteristic (AUROC) curve. To further elucidate the relationship between variables and geriatric depression, logistic regression was subsequently applied. Results: Our findings accentuated that sleep patterns emerged as the paramount determinants influencing the onset of depression in the elderly. Relationships with offspring ranked as the second most significant determinant, only surpassed by sleep habits. A negative correlation was observed between sleep patterns (Odds Ratio [OR]: 0.78, 95 % Confidence Interval [CI]: 0.75-0.81, P < 0.01), communication with offspring (OR: 0.86, 95 % CI: 0.82-0.90, P < 0.01), and the prevalence of depressive symptoms. Among the evaluated models, the k-Near Neighbor algorithm demonstrated commendable discriminative power. However, it was the Random Forest algorithm that manifested unparalleled discriminative prowess and precision, establishing itself as the most efficacious classifier. Conclusion: Prolonging the duration of nocturnal sleep, and elevating the frequency of communication with offspring have been identified as measures conducive to mitigating the onset of geriatric depression.

Diabetic retinopathy (DR) is a severe ocular complication of diabetes that can lead to vision damage and even blindness. Currently, traditional deep convolutional neural networks (CNNs) used for DR grading tasks face two primary challenges: (1) insensitivity to minority classes due to imbalanced data distribution, and (2) neglecting the relationship between the left and right eyes by utilizing the fundus image of only one eye for training without differentiating between them. To tackle these challenges, we proposed the DRGCNN (DR Grading CNN) model. To solve the problem caused by imbalanced data distribution, our model adopts a more balanced strategy by allocating an equal number of channels to feature maps representing various DR categories. Furthermore, we introduce a CAM-EfficientNetV2-M encoder dedicated to encoding input retinal fundus images for feature vector generation. The number of parameters of our encoder is 52.88, which is less than RegNet_y_16gf (80.57M) and EfficientNetB7 (63.79M), but the corresponding kappa value is higher. Additionally, in order to take advantage of the binocular relationship, we input fundus retinal images from both eyes of the patient into the network for features fusion during the training phase. We achieved a kappa value of 86.62% on the EyePACS dataset and 86.16% on the Messidor-2 dataset. Experimental results on these representative datasets for diabetic retinopathy (DR) demonstrate the exceptional performance of our DRGCNN model, establishing it as a highly competitive intelligent classification model in the field of DR. The code is available for use at https://github.com/Fat-Hai/DRGCNN.