In the task of pneumonia recognition from chest X-ray (CXR) images, many excellent models based on convolutional neural networks have been proposed. However, the majority of these models overlook the contextual and positional information of the image, leading to difficulties in accurately identifying the location of lesions. To address these limitations, we present a novel convolutional model with a Transformer-style architecture called CEGNet, which incorporates both position attention and spatial attention mechanisms.To enable the convolutional block to capture the global contextual information of the image, we propose a new module called Convolution Transformer (CTf), which simplifies the computation of similarity score matrices in the self-attention mechanism. To emphasize the importance of image location information, we introduce an Efficient Position Attention (EPA) module. The EPA module decomposes the channels along specific directions into two 1D encodings, capturing precise position information along those directions and applying it to the original image.Additionally, we propose a lightweight spatial attention module called Gelu Spatial Attention (GSA), which uses three consecutive 3×3 convolutions and the GELU activation function to focus on relevant spatial information in the image while reducing the number of parameters. The CTf, EPA, and GSA modules complement each other in capturing different aspects of image information, thereby enhancing the model's ability to recognize pneumonia.We explore the internal structure of the CEGNet model on the COVID-19 Radiography Dataset V5 and achieve a final accuracy of 95.805%, precision of 95.814%, recall of 95.805%, and F1-Score of 95.796% on this dataset. Furthermore, we validate the CEGNet model on the ChestXRay2017 dataset, where it also achieves excellent performance.

Coronavirus disease2019 (COVID-19)has become a global pandemic. Many recognition approaches based on convolutional neural networks have been proposed for COVID-19 chest X-ray images. However, only a few of them make good use of the potential inter- and intra-relationships of feature maps. Considering the limitation mentioned above, this paper proposes an attention-based convolutional neural network, called PCXRNet, for diagnosis of pneumonia using chest X-ray images. To utilize the information from the channels of the feature maps, we added a novel condense attention module (CDSE) that comprised of two steps: condensation step and squeeze-excitation step. Unlike traditional channel attention modules, CDSE first downsamples the feature map channel by channel to condense the information, followed by the squeeze-excitation step, in which the channel weights are calculated. To make the model pay more attention to informative spatial parts in every feature map, we proposed a multi-convolution spatial attention module (MCSA). It reduces the number of parameters and introduces more nonlinearity. The CDSE and MCSA complement each other in series to tackle the problem of redundancy in feature maps and provide useful information from and between feature maps. We used the ChestXRay2017 dataset to explore the internal structure of PCXRNet, and the proposed network was applied to COVID-19 diagnosis. As a result, the network achieves an accuracy of 94.619%, recall of 94.753%, precision of 95.286%, and F1-score of 94.996% on the COVID-19 dataset.

Thyroid cancer is extremely common in the population and its incidence has gradually increased in recent years. In order to help doctors diagnose thyroid nodules, many computer diagnosis algorithms have been proposed, but most of these algorithms only focus on improving the accuracy of recognition of benign and malignant thyroid nodules, the neural network model established is relatively complex, which will lead to longer inference time. In addition, due to the limitation of hardware, the model with a large number of parameters can easily make the relevant equipment unable to function normally. To solve the above problems, this paper proposes a lightweight residual network, EDSResNet, by improving ResNet-34. In the experimental part, Vgg-16, AlexNet, MobileNet_v2, ResNet, ShuffleNet_v2_x2.0, and DenseNet-121 are introduced for comparison, and the results show that the proposed EDSResNet improves 1.1% in accuracy compared with the ResNet-34, and the parameter quantity of EDSResNet is only 6.6% of that before the improvement, the number of floating point operations (Flops) is 10% of that before the improvement. Compared with another lightweight network, MobileNet_v2, EDSResNet has $8\times 10^{5}$ fewer parameters, and the accuracy, sensitivity and specificity are higher by 1.7%, 2.0% and 0.8%, respectively. After comparing with all networks, it can be found that the EDSResNet proposed in this paper has an excellent classification effect on the ultrasound image dataset of thyroid nodules while having the smallest number of parameters.

Early rehabilitation exercises after elbow fracture treatment can affect the degree of joint function recovery. Due to the different levels of endurance and pain tolerance of patients, blind training not only makes patients to suffer pain, but also tends to cause postoperative complications. Therefore, this paper introduces a wearable assistive elbow rehabilitation training device that can quantify the patient's recovery level. The device uses pressure sensors to collect the pressure value of the patient's arm when squeezing the muscles, tri-axial gyroscopes and tri-axial accelerometers to perceive the angular velocity and acceleration of the patient's arm during the rehabilitation exercise. The collected data are processed by an internal processor and presented in a touch screen, which in turn provides a reference for the physician to develop a rehabilitation training program. In addition, the device also provides voice alerts when the patient's rehabilitation training reaches an unreasonable level or range of motion, helping the patient to carry out quantized rehabilitation training and promote normal fracture healing.

Rib fractures are one of the most common clinical indicators of the severity of trauma, so accurate and rapid identification of all rib fracture areas and determination of the severity of the patient's trauma is essential for clinicians to follow up. In this paper, we propose an automatic rib fracture recognition model based on nnU-Net and Densenet. In the first stage, we train a deep learning segmentation model that can generate candidate rib fracture regions; in the second stage, we classify the segmented candidate fracture regions and determine whether they are fractures based on the classification results. The experimental results show that the two-stage rib fracture automatic recognition model proposed in this paper reduces the false positive and false negative rates of rib fracture detection, improves the accuracy of rib fracture recognition, and can assist physicians in accurately and rapidly identifying rib fracture areas.