Vision Transformer has made breakthrough in medical image diagnosis. However, its limited feature representation capability and oversized parameter may make computation inefficient and deployment for downstream tasks difficult. To address these issues, a novel lightweight knowledge distillation framework with adaptive feature selection is proposed. A multi-scale kernel alignment method is employed to quantify the differences between different feature layers, adaptively select discriminative feature layers, and apply a deep alignment strategy with soft-weighted fusion to guide the student model in knowledge distillation. A guided matrix is used to extract attention features and hidden state features from the teacher model, filter out redundant features, and introduce learnable low-rank increments to enhance the convergence of the student model, enabling efficient knowledge transfer. An entropy-based sparse pruning strategy dynamically adjusts pruning intensity according to the information entropy of the parameter matrix, increasing inference speed while maintaining higher prediction performance and better computational efficiency. Using 1764 Computed Tomography scans of patients with chronic obstructive pulmonary disease from Zhongshan Hospital, and 1414 Computed Tomography scans of patients with patellar fracture from Shanghai Sixth People's Hospital, three models of different scales were distilled and trained. After 3-fold cross-validation, the smallest model achieved accuracy of 78.83% and 78.52%, recall of 79.27% and 78.80%, and F1 scores of 78.92% and 78.58%, respectively. Compared with state-of-the-art lightweight models, the proposed model enjoys higher prediction accuracy and generalization with fewer parameters. Furthermore, key clinical practicality metrics of the proposed model are analyzed to facilitate diagnosis.

Pulse wave prediction of hypertensive risk stratification by deep learning often comes with low prediction accuracy and poor real-time performance due to the low distinguishability of pulse wave features and high algorithm complexity. We propose a time-frequency feature fusion model. The model transforms the time-domain pulse wave into a 144-dimensional time-domain and a 36-dimensional frequency-domain feature map, which are then fused in parallel connection-superposition to increase information volume and dimensionality, thereby enhancing the distinguishability of pulse wave features. Multiple small-scale dilated convolution kernels are stacked to replace large-scale standard convolutions, reducing the effects of "receptive field center bias" and model parameters, while enhancing global feature perception capabilities. The structure reparameterization technique is introduced into dilated convolutions, utilizing depthwise separable convolutions to replace largescale dilated convolutions, and channel shuffling operation is introduced to increase cross-channel information exchange, further reducing model redundancy and improving prediction accuracy and generalization performance. The Clinical pulse diagnosis data provided by Longhua Hospital affiliated with Shanghai University of Traditional Chinese Medicine and the Integrated Traditional Chinese and Western Medicine Hospital were used. After 5-fold cross-validation, the evaluation metrics F1-score, Accuracy, Recall, Specificity, and AUC were 90.87 %, 92.42 %, 91.90 %, 95.78 %, and 98.91 %, respectively. Compared to the typical models, the proposed method shows higher classification accuracy and generalization performance. Additionally, we investigate the correlation between pulse wave time-domain and frequency-domain features and hypertensive risk stratification based on random forests and effective receptive field weight proportion algorithms, delving into the potential risk factors and providing effective references for clinical diagnosis of hypertension.

Chronic Obstructive Pulmonary Disease (COPD) poses a significant challenge to global health systems, but timely detection plays a vital role in mitigating disease progression and enhancing patient outcomes. However, with complex structures the existing prediction models cannot extract key features efficiently, which reduces the model’s convergence and inference capabilities. We proposed a lightweight model for COPD diagnosis. The model utilized techniques such as channel shuffling, horizontal and vertical coordinate encoding, and multi-scale group convolution to construct a two-branch separable convolution module, which reduced the computation and improved the prediction accuracy. A lightweight hybrid spatial-channel attention mechanism was used to enhance the feature extraction capability, reduce the algorithm complexity, and further improve the inference speed. We also introduced the “Sigmoid cross-generation” loss function which dynamically adjusted the loss weights in different training phases, accelerated the training convergence, and enhanced generalization performance. Furthermore, in the clinical practicality analysis, our model achieved higher net benefits of the decision curve analysis across specific thresholds, and a stronger alignment of the calibration curve between predicted probabilities and actual outcomes. Experiments were conducted on 13,991 CT scans from 90 COPD cases at Zhongshan Hospital, Fudan University. Our model achieved 5-fold cross-validation metrics of 93.69% Accuracy, 93.73% Precision, 93.66% Recall, and 93.65% F1-score, highlighting the superior accuracy and generalization of the proposed model from the other lightweight architectures.

Deep learning-based computed tomography (CT) image prediction models often suffer from poor local feature perception, severe class imbalance, and a scarcity of high-quality labeled data. Existing fine-tuning methods are sample-inefficient and prone to overfitting. To address these issues, this study proposes a Qwen3-VL vision-language model based on adaptive semi-supervised learning. A novel data augmentation strategy for minority classes is introduced, utilizing diffusion models to learn the distribution of minority-class data and generate new samples. Perceptual similarity and generation quality are evaluated to ensure visual fidelity, while two-level feature selection is performed based on local density to increase the number of high-quality samples in the feature space. A fine-tuning method based on log-sum regularization is proposed, which decomposes pre-trained weights into direction and magnitude components, guiding the magnitude along specific directions to enhance the perception of fine-grained features. The multi-view prediction problem is transformed into a mass function using evidence theory, quantifying uncertainty and generating high-confidence pseudo-labels to improve prediction accuracy and generalization performance. Experiments are conducted on 1 980 CT samples of interstitial lung disease related to rheumatoid arthritis from Guang’anmen Hospital, China Academy of Chinese Medical Sciences, using three-fold cross-validation. The proposed model achieves accuracy, recall, and F1-score of 89.85

Machine learning applied to the clinical prediction of breast cancer mainly deals with a single view feature, so class imbalance, low classification accuracy and poor generalization performance may occur. Addressing the aforementioned problems, this study proposes a semi-supervised learning framework for predicting breast cancer metastasis. An integrated over-sampling and under-sampling approach is employed to construct multiple diverse weak classifiers, aiming to mitigate prediction bias towards the majority class and enhance generalization performance. By improving the classical Dempster–Shafer theory, the credibility of evidence sources is dynamically adjusted based on the conflict between them, thereby enhancing the accuracy of multi-source feature fusion. This improved DS theory is integrated with the Tri-training collaborative training strategy. Credibility thresholds are used to generate pseudo-labels from unlabeled data, which expands the labeled dataset, reduces prediction conflicts among multiple models, and improves both prediction accuracy and generalization. Furthermore, the SHAP interpretability method is adopted for feature importance analysis, aiming to identify potential risk factors and provide technical support for the clinical diagnosis of breast cancer. The 304 clinical cases of breast cancer are from Shanghai Pudong New Area Gongli Hospital. The Accuracy, Precision, Recall, F1-score, and AUC on threefold cross-validation are 94.41, 94.44, 93.51, 93.88 and 93.5

3D-based medical image segmentation, offering enhanced spatial information compared to 2D slice-based methods, encounters challenges arising from factors such as a restricted clinical sample size, imbalanced foreground-background pixel distribution, and suboptimal generalization performance. To address these challenges, we propose a lightweight segmentation model tailored to 3D medical images. Employing the K-means algorithm, our approach efficiently extracts the Region of Interest (ROI) from medical images, facilitating lung area segmentation while minimizing interference from background pixels. We address the risk of model over- fitting by adopting the Focal loss in conjunction with the Dice coefficient as our loss function. Feature extraction capabilities are bolstered through the incorporation of a parallel attention mechanism at skip connections, aiming to enhance the representation of both shallow and deep layers. Moreover, we optimize computational efficiency and memory utilization by substituting 3 x 3 convolutions with depth-wise separable convolutions and integrating residual connections for improved gradient propagation. The introduction of Ghost-inspired 1 x 1 convolution ensures consistent feature dimensions before and after residual connections. Experimental evaluation, conducted on a dataset comprising 199 COVID-19-Seg cases through 5-fold cross-validation, underscores the superior performance of our proposed model. Evaluation metrics, including Average Surface Distance (ASD), accuracy, sensitivity, Dice coefficient, and Intersection over Union (IOU) accuracy, yield values of 19.880, 99.90 %, 58.90 %, 56.10 %, and 41.00 %, respectively. In comparison to the other state-of-the-art segmentation models, our approach achieves heightened segmentation accuracy and generalization performance while incurring only a marginal increase in parameters and computational complexity.

The application of machine learning algorithms in clinical disease diagnosis has significantly improved the accuracy and efficiency of clinical predictions. However, many clinical datasets suffer from a severe class imbalance, which causes the predictions to be biased towards the majority class. Traditional sampling methods often alter the original distribution of samples, which negatively impacts the prediction accuracy and generalization ability of models. In this paper, we propose a cascaded proportion-balanced ensemble learning model for prediction of disease activity in rheumatoid arthritis, which consists of a symmetric three-part structure, with each part being a cascaded structure made up of a base classifier and a sampling model. Through cascaded proportion sampling, the majority class samples are randomly deleted and minority class samples are replicated in proportion, gradually balancing the training sample distribution for the base classifiers. Then, the sample sizes of each class are adjusted in reverse to achieve overall sample balance for the cascaded structure. We use an asymmetric Beta distribution function to fit the weight changes of base classifiers, controlling the curve shape with two variables to dynamically adjust the weight distribution of each base model, ensuring the symmetry and effectiveness of the weighted ensemble model. The Bayesian optimization algorithm is employed to dynamically search for the optimal hyperparameters of the classification model, further enhancing prediction accuracy and generalization ability. We evaluate the model using 5-fold cross-validation on a dataset of 10 514 rheumatoid arthritis samples from the Guang’anmen Hospital, China Academy of Chinese Medical Sciences. The evaluation metrics show GMarea of 0.60, Precision of 0.78, and AUC of 0.83. Compared to various typical models, the proposed model demonstrates better prediction accuracy and generalization performance.

In recent years, artificial intelligence has been applied to 3D COVID-19 medical image diagnosis, which reduces detection costs and missed diagnosis rates with higher predictive accuracy, and diagnostic efficiency. However, the limited size and low quality of clinical 3D medical image samples have hindered the segmentation performance of 3D models. Therefore, we propose a 3D medical image segmentation model based on semi-supervised learning using co-training. Multi-view and multi-modal images are generated using spatial flipping and windowing techniques to enhance the spatial diversity of 3D image samples. A pseudo label generation module based on confidence-weights is employed to generate reliable pseudo labels for non-annotated data, thereby increasing the sample size and reducing overfitting. The proposed approach utilizes a three-stage training process: firstly, training a single network based on annotated data; secondly, incorporating non-annotated data to train a dual-modal network and generate pseudo labels; finally, jointly training six models in three dimensions using both annotated and pseudo labels generated from multi-view and multi-modal images, aiming to enhance segmentation accuracy and generalization performance. Additionally, a consistency regularization loss is applied to reduce noises and accelerate convergence of the training. Moreover, a heatmap visualization method is employed to focus on the attention of features at each stage of training, providing effective reference for clinical diagnosis. Experiments were conducted on an open dataset of 3D COVID-19 CT samples and a non-annotated dataset from TCIA, including 771 NIFTI-format CT images from 661 COVID-19 patients. The results of 5-fold cross-validation show that the proposed model achieves a segmentation accuracy of Dice=73.30%, ASD=10.633, Sensitivity=63.00%, and Specificity=99.60%. Compared to various typical semi-supervised learning 3D segmentation models, it demonstrates better segmentation accuracy and generalization performance.

The application of deep learning to the classification of pulse waves in Traditional Chinese Medicine (TCM) related to hypertensive target organ damage (TOD) is hindered by challenges such as low classification accuracy and inadequate generalization performance. To address these challenges, we introduce a lightweight transfer learning model named MobileNetV2SCP. This model transforms time-domain pulse waves into 36-dimensional frequency-domain waveform feature maps and establishes a dedicated pre-training network based on these maps to enhance the learning capability for small samples. To improve global feature correlation, we incorporate a novel fusion attention mechanism (SAS) into the inverted residual structure, along with the utilization of 3×3 convolutional layers and BatchNorm layers to mitigate model overfitting. The proposed model is evaluated using cross-validation results from 805 cases of pulse waves associated with hypertensive TOD. The assessment metrics, including Accuracy (92.74%), F1-score (91.47%), and Area Under Curve (AUC) (97.12%), demonstrate superior classification accuracy and generalization performance compared to various state-of-the-art models. Furthermore, this study investigates the correlations between time-domain and frequency-domain features in pulse waves and their classification in hypertensive TOD. It analyzes key factors influencing pulse wave classification, providing valuable insights for the clinical diagnosis of TOD.

Convolutional neural networks (CNNs) can be used for clinical medical image segmentation to improve detection efficiency and accuracy. However, because of the fixed size of convolution kernel and receptive field for existing CNN model, some associated pixel features are ignored and segmentation performance is impaired. Therefore, we propose an effective image segmentation model, COPANet, which uses RepVGG as the backbone and replaces the standard convolution with dilated convolution of multiple kernels to increase the receptive field size, so that COPANet can make full use of pixel correlations and acquire contexture semantic information. We also build skip-connections between global and local features, where a parallel attention (PANet) is employed to extract important location information in the downsampling. PANet is a fused mechanism integrating channel attention with spatial attention in parallel, which can fully extract more global information. We also apply weighted combined loss to reduce the effect of class imbalance of foreground and background pixels on segmentation performance and speed up convergence. In addition, we conduct experiments on 480 cases of clinical CT sinus from Shanghai Tongji Hospital and 236 cases of CT patella fracture from Shanghai Sixth People's Hospital. The evaluation indexes after 5-Fold cross-validation are as follows: Precision is 92.76% and 96.55%, Recall is 88.25% and 57.58%, Specificity is 99.85% and 96.55%, IOU is 91.08% and 77.38%, Hausdorff Distance is 2.8135 and 11.1879, respectively. Compared with the state-of-the-art models, COPANet has higher segmentation accuracy and better generalization performance, which can assist the clinical diagnosis.

Objective In the application of machine learning to the prediction of hypertension, many factors have seriously affected the classification accuracy and generalization performance. We propose a pulse wave classification model based on multi-feature fusion for accuracy prediction of hypertension. Methods and Materials We propose an ensemble under-sampling model with dynamic weights to decrease the influence of class imbalance on classification, further to automatically classify of hypertension on inquiry diagnosis. We also build a deep learning model based on hybrid attention mechanism, which transforms pulse waves to feature maps for extraction of in-depth features, so as to automatically classify hypertension on pulse diagnosis. We build the multi-feature fusion model based on dynamic Dempster/Shafer (DS) theory combining inquiry diagnosis and pulse diagnosis to enhance fault tolerance of prediction for multiple classifiers. In addition, this study calculates feature importance ranking of scale features on inquiry diagnosis and temporal and frequency-domain features on pulse diagnosis. Results The accuracy, sensitivity, specificity, F1-score and G-mean after 5-fold cross-validation were 94.08%, 93.43%, 96.86%, 93.45% and 95.12%, respectively, based on the hypertensive samples of 409 cases from Longhua Hospital affiliated to Shanghai University of Traditional Chinese Medicine and Hospital of Integrated Traditional Chinese and Western Medicine. We find the key factors influencing hypertensive classification accuracy, so as to assist in the prevention and clinical diagnosis of hypertension. Conclusion Compared with the state-of-the-art models, the multi-feature fusion model effectively utilizes the patients’ correlated multimodal features, and has higher classification accuracy and generalization performance.

Background: The diagnosis of allergic rhinitis (AR) primarily relies on symptoms and laboratory examinations. Due to limitations in outpatient settings, certain tests such as nasal provocation tests and nasal secretion smear examinations are not routinely conducted. Although there are clear diagnostic criteria, an accurate diagnosis still requires the expertise of an experienced doctor, considering the patient’s medical history and conducting examinations. However, differences in physician knowledge and limitations of examination methods can result in variations in diagnosis. Objective: Artificial intelligence is a significant outcome of the rapid advancement in computer technology today. This study aims to present an intelligent diagnosis and detection method based on ensemble learning for AR. Method: We conducted a study on AR cases and 7 other diseases exhibiting similar symptoms, including rhinosinusitis, chronic rhinitis, upper respiratory tract infection, etc. Clinical data, encompassing medical history, clinical symptoms, allergen detection, and imaging, was collected. To develop an effective classifier, multiple models were employed to train on the same batch of data. By utilizing ensemble learning algorithms, we obtained the final ensemble classifier known as adaptive random forest-out of bag-easy ensemble (ARF-OOBEE). In order to perform comparative experiments, we selected 5 commonly used machine learning classification algorithms: Naive Bayes, support vector machine, logistic regression, multilayer perceptron, deep forest (GC Forest), and extreme gradient boosting (XGBoost).To evaluate the prediction performance of AR samples, various parameters such as precision, sensitivity, specificity, G-mean, F1-score, and area under the curve (AUC) of the receiver operating characteristic curve were jointly employed as evaluation indicators. Results: We compared 7 classification models, including probability models, tree models, linear models, ensemble models, and neural network models. The ensemble classification algorithms, namely ARF-OOBEE and GC Forest, outperformed the other algorithms in terms of the comprehensive classification evaluation index. The accuracy of G-mean and AUC parameters improved by nearly 2% when compared to the other algorithms. Moreover, these ensemble classifiers exhibited excellent performance in handling large-scale data and unbalanced samples. Conclusion: The ARF-OOBEE ensemble learning model demonstrates strong generalization performance and comprehensive classification abilities, making it suitable for effective application in auxiliary AR diagnosis.

Common clinical rhinitis is characterized by different types of cases and class imbalance. Its prediction belongs to multiple output classification. Low recognition rate and poor generalization performance often occur for minority class. Therefore, we propose a novel integrated classification model, ARF-OOBEE, which transforms the multi-output classification to multi-label classification and multi-class classification. The multi-label classifier automatically adjusts the number and depth of integrated forest learners according to the imbalance ratio of single class label in a subset. It can effectively reduce the impact of class imbalance on classification and improve prediction performance of both majority or minority class concurrently. Also, we build a multi-class classification based on out-of-bag Extra-Tree to accomplish finer classification for the predicted labels. In addition, we calculate the feature importance for rhinitis on the grounds of the purity of nodes in decision-making tree inside Random Forest and study the correlation between rhinitis features. We conduct 12 folds cross-validation experiments on 461 cases of clinical rhinitis. The outcomes show that the evaluation indicators of ARF-OOBEE, such as Sensitivity, Specificity, Accuracy, F1-Score, AUC, and G-Mean are 74.9%,86.5%,92.0%,78.3%,95.3%, and 79.9%, respectively. In comparison to the other methods, ARF-OOBEE has better evaluation indicator and is more effective for the early clinical diagnosis of rhinitis.

With respect to less efficiency and low accuracy of predicting on hypertensive target organ damage, this article proposes a fusion prediction model combining pulse-taking with inquiry diagnosis of traditional Chinese medicine to accomplish the efficient and non-invasive diagnosis. Regarding the class imbalance of inquiry diagnosis samples, an Eliminated random forest algorithm is proposed to select efficient features and reduce the impact of class imbalance on classification performance via cluster-based under-sampling algorithm. As to low discriminability of hypertensive time-domain pulse wave samples, time-domain pulse wave is transformed to the frequency-domain MFCC feature maps, and fuse feature maps of inquiry diagnosis scale for predicting hypertension target organ damage. In the article, the clinical 608 cases of hypertensive target organ damage are from Longhua Hospital affiliated to Shanghai University of Chinese Medicine and Hospital of Integrated Traditional Chinese and Western Medicine concerning pulse-taking and inquiry diagnosis. The evaluation indicators of 5-Fold crossvalidation classification, i.e. F1-score, Accuracy, Precision, Sensitivity, AUC, are 97.31%, 98.72%, 97.71%, 97.04%, 99.13% respectively, which are higher than those of the other typical models. In addition, this article also studies the correlation between classification of pulse-taking or inquiry diagnosis and its features, and analyzes the feature importance ranking on pulse-taking and inquiry diagnosis, which aids clinicians to seek the occurrence mechanisms of hypertensive target organ damage, and find the effective measurements for timely prevention and treatment.

In order to solve the problems of low segmentation accuracy and pixel rasterization of segmentation results of psoriasis images, an improved U-net method for medical image segmentation is proposed in this paper.A contrast limited adaptive histogram equalization（CLAHE） algorithm is used to preprocess the psoriasis image.The jump connected Inception-Res A block is introduced to replace the feature extraction unit of the classic Unet model for image segmentation. The Markov random field is used for post-processing to further improve the accuracy of segmentation. For clinical psoriasis samples in Changzheng Hospital, the experimental results show that the Dice coefficient and Jaccard coefficient of the improved U-net model are 88.4% and 79.5%, respectively.Compared with the traditional U-net model, this model has higher segmentation accuracy and better generalization performance, and reduces the rasterization of medical image segmentation results.

Purpose:To explore the typical CT signs of strangulated small bowel obstruction,and to screen the important signs of surgical treatment by using the predictive modeling of small bowel obstruction and its characteristic correlation study.Methods:One hundred patients diagnosed with small bowel obstruction in our hospital from January 2016 to October 2021 were divided into strangulation small bowel obstruction group(n=20)and non-strangulation small bowel obstruction group(n=80)according to surgical results.Imaging and clinical data of all patients were collected.The distribution difference of different CT signs between the two groups was analyzed by predictive modeling of small bowel obstruction and stenosis.Combined diagnosis was made according to the difference and logistic regression,and the CT diagnosis combination with high diagnostic value was screened.Results:Among 100 patients,there were statistical significant differences in the distribution of mesenteric edema,intestinal wall thickening,and increased or decreased intestinal wall density between the two groups(P＜0.05).Conclusion:The prediction modeling method of small bowel obstruction and its characteristic correlation study have certain diagnostic value in the diagnosis of strangulated small bowel obstruction by CT plain scan.

Medical image segmentation is the key research of precision medicine. The existing models often ignore some important pixel features and fail to effectively extract global correlation features, which causes poor performance of segmentation. In this paper, we propose TSE DeepLab, which retains the original atrous convolution for extraction of local feature on the basis of DeepLabv3 framework, converts the feature maps after backbone into visual tokens, and further feeds them into Transformer module to enhance the ability of global feature extraction. At the same time, squeeze and excitation components are added to sort the importance of channels after Transformer module, so that the model pays attention to the important pixel features of each channel. In this paper, we apply 5-fold cross-validation to study the clinical sinus instances of Shanghai Tongji Hospital affiliated to Tongji University and the patellar fracture instances of the Sixth People's Hospital affiliated to Shanghai Jiao Tong University. The average of evaluation measures achieves Accuracy of 99.74%, Precision of 93.67%, IOU of 88.10%, Specificity of 99.87%, Fl-score of 93.63%, Sensitivity of 93.82% on sinus dataset and Accuracy of 99.53%, Precision of 85.64%, IOU of 78.47%, Specificity of 99.72%, Fl-score of 87.15%, Sensitivity of 89.95% on patellar fracture dataset. Compared with various typical segmentation models, the proposed model attains better segmentation accuracy and generalization performance, and has better reference value for clinical medical diagnosis.

为解决临床医学量表数据类别不均衡容易对模型产生影响,以及在处理量表数据任务时深度学习框架性能难以媲美传统机器学习方法问题,提出了一种基于级联欠采样的Transformer网络模型(layer by layer Trans-former,LLT).LLT通过级联欠采样方法对多数类数据逐层删减,实现数据类别平衡,降低数据类别不均衡对分类器的影响,并利用注意力机制对输入数据的特征进行相关性评估实现特征选择,细化特征提取能力,改善模型性能.采用类风湿关节炎(RA)数据作为测试样本,实验证明,在不改变样本分布的情况下,提出的级联欠采样方法对少数类别的识别率增加了 6.1％,与常用的NEARMISS和ADASYN相比,分别高出1.4％和10.4％;LLT在RA量表数据的准确率和F1-score指标上达到了 72.6％和71.5％,AUC值为0.89,mAP值为0.79,性能超过目前RF、XGBoost和GBDT等主流量表数据分类模型.最后对模型过程进行可视化,分析了影响RA的特征,对RA临床诊断具有较好的指导意义.

Rhinitis is a common chronic inflammation of the upper respiratory tract with multiple symptoms and signs. The clinical classification of rhinitis is a multi-label classification, characterized by high dimensionality, poor correlation and class imbalance. Low recognition rate and poor generalization performance often occur for minority class instances. Therefore, this paper proposes a Feature-Block classification model, MBLCC based on Label-Links Classifier Chain. We apply kernel density estimation and Deborah Hellinger distance to partition allergic rhinitis instance with similar characteristics, calculate the correlation matrix of label characteristics, build an ordered classification chain for each block, and integrate the predictions of each block classifiers by evidence theory as the outcome. The cross-validation experiments conducted on 2231 cases of clinical rhinitis show that the evaluation indicators of MBLCC, i.e. sensitivity, specificity, accuracy, F1-score, and G-Mean, are 91.80 %, 96.8 %, 96.9 %, 0.925, and 0.941 respectively. In comparison with the other baselines, MBLCC achieves better generalization performance and is more effective and rapid in early clinical diagnosis of rhinitis. In addition, we calculate the feature importance ranking for rhinitis features via Label-Links Classifier Chain on the grounds of the purity of nodes in decision-making tree inside Random Forest and study the correlation between rhinitis features and classification that can provides the reference for the clinical rhinitis diagnosis.

Purpose At present, though the application of Convolution Neural Network (CNN) to detect COVID-19 infection significantly enhance the detection performance and efficiency, it often causes low sensitivity and poor generalization performance. Methods In this article, an effective CNN, CrodenseNet is proposed for COVID-19 detection. CrodenseNet consists of two parallel DenseNet Blocks, each of which contains dilated convolutions with different expansion scales and traditional convolutions. We employ cross-dense connections and one-sided soft thresholding to the layers for filtering of noise-related features, and increase information interaction of local and global features. Results Cross-validation experiments on COVID-19x dataset shows that via CrodenseNet the COVID-19 detection attains the precision of 0.967 +/- 0.010, recall of 0.967 +/- 0.010, F1-score of 0.973 +/- 0.005, AP (area under P-R curve) of 0.991 +/- 0.002, and AUC (area under ROC curve) of 0.996 +/- 0.001. Conclusion CrodenseNet outperforms a variety of state-of-the-art models in terms of evaluation metrics so it assists clinicians to prompt diagnosis of COVID-19 infection.