Accurate classifications of colorectal cancer (CRC) lymph node metastasis (LNM) could assist radiologists in increasing the diagnostic accuracy and help surgeons establish a correct surgical plan. This study aims to present an efficient pipeline with deep transfer learning for CRC LNM classification. Hence, 11 deep pre-trained models have been investigated on a CRC LN dataset. The dataset of this experiment is from Harbin Medical University Cancer Hospital. This dataset contains samples of 619 patients. Among these samples, 312 were positive and 307 were negative. In addition, datasets with different dimensions and various training epochs were also studied to ascertain the minimum training dataset and training times. In order to improve the interpretability of the model classification performance, a visual convolution layer feature map was first established to compute the similarity distance between the feature map and original data. The experimental results revealed that resnet_152 was the best deep pre-trained model for the classification of CRC LNM, with an accuracy of 97.2%, with 600 raw data samples being the minimum dimension of a dataset and 30 epochs the minimum training times in the CRC LNM classification. This study suggests that the proposed deep transfer learning pipeline could classify the CRC LNM with high efficiency, without requiring sophisticated computational knowledge for radiologists.

In a complex background, insulator fault is the main factor behind transmission accidents. With the wide application of unmanned aerial vehicle (UAV) photography, digital image recognition technology has been further developed to detect the position and fault of insulators. There are two mainstream methods based on deep learning: the first is the “two-stage” example for a region convolutional neural network and the second is the “one-stage” example such as a single-shot multibox detector (SSD), both of which pose many difficulties and challenges. However, due to the complex background and various types of insulators, few researchers apply the “two-stage” method for the detection of insulator faults in aerial images. Moreover, the detection performance of “one-stage” methods is poor for small targets because of the smaller scope of vision and lower accuracy in target detection. In this article, the authors propose an accurate and real-time method for small object detection, an example for insulator location, and its fault inspection based on a mixed-grouped fire single-shot multibox detector (MGFSSD). Based on SSD and deconvolutional single-shot detector (DSSD) networks, the MGFSSD algorithm solves the problems of inaccurate recognition in small objects of the SSD and complex structure and long running time of the DSSD. To resolve the problems of some target repeated detection and small-target missing detection of the original SSD, the authors describe how to design an effective and lightweight feature fusion module to improve the performance of traditional SSDs so that the classifier network can take full advantage of the relationship between the pyramid layer features without changing the base network closest to the input data. The data processing results show that the method can effectively detect insulator faults. The average detection accuracy of insulator faults is 92.4% and the average recall rate is 91.2%.

Cardiovascular disease is one of the major diseases that threaten the human health. But the existing electrocardiograph (ECG) monitoring system has many limitations in practical application. In order to monitor ECG in real time, a portable ECG monitoring system based on the Android platform is developed to meet the needs of the public. The system uses BMD101 ECG chip to collect and process ECG signals in the Android system, where data storage and waveform display of ECG data can be realized. The Bluetooth HC-07 module is used for ECG data transmission. The abnormal ECG can be judged by P wave, QRS bandwidth, and RR interval. If abnormal ECG is found, an early warning mechanism will be activated to locate the user’s location in real time and send preset short messages, so that the user can get timely treatment, avoiding dangerous occurrence. The monitoring system is convenient and portable, which brings great convenie to the life of ordinary cardiovascular users.

In traditional CBCT guided radiotherapy, the conventional process is to scan a planned CT image of the patient before treatment, and use the CT image to prepare a treatment plan for the patient, and calculate the radiation dose with the electronic density information of the CT image to obtain the radiation dose that the patient needs to receive. Because CT images cannot be directly used to calculate the amount of data, in order to solve the problem of CT image attenuation corresponding to MRI image synthesis, the deep convolution network model is used to map the CT image to the MRI image, input the CT image, and synthesize the corresponding MRI image with the convolution network model in this article. The synthetic MRI image can be used for the same mode registration with the patient’s positioning MRI image, so as to solve the problem of inaccurate cross-membrane registration. The multi-mode synthesis and transformation of CT/MRI images have been realized. Experiments have proved that the method presented in this article is beneficial to reducing the radiation dose of patients, enabling patients to receive more accurate radiotherapy, so that the tumor part can be irradiated as much as possible and the normal tissues around the tumor can be irradiated less, so as to improve the therapeutic effect of tumor patients.

Automatic medical image segmentation effectively aids in stroke diagnosis and treatment. In this article, an improved U-Net neural network for auxiliary diagnosis of intracerebral hemorrhage is proposed, which can realize the automatic segmentation of hemorrhage from brain CT images. The pixels of brain CT images are first clustered into four classes: gray matter, white matter, cerebrospinal fluid, and hemorrhage by fuzzy c-means (FCM) clustering, followed by the removal of the skull by morphological imaging, and finally an improved U-Net neural network model is proposed to automatically segment hemorrhages from the brain CT images. Experiment results showed that the objective function of binary cross-entropy was better than dice loss and focal loss for the proposed method. Its dice similarity coefficient reached 0.860 ± 0.031, which was better than the methods of white matter FCM clustering and multipath context generation adversarial networking. This improved method dramatically enhanced the accuracy of segmentation for intracerebral hemorrhage.

In the field of single-image super-resolution (SISR) research, neural networks and deep learning methods are gradually being widely used by researchers. Over time, the fields of application have expanded in scope. The SISR method is also applied in the field of intelligent satellite imagery. In recent years, research applications based on intelligent satellite images have mostly focused on imaging, classification, and segmentation. They have rarely been used in actual observation problems. This article proposes a new intelligent neural network model, the Laplacian pyramid residual dense network, for the super-resolution of hyperspectral satellite medical geographic small-targets. This study proceeds in three steps. First, the three-layer Laplacian pyramid structure is designed to increase the depth of the image at the feature extraction stage. Second, the residual mode is improved and updated; a new residual block is proposed for constructing the residual dense network to enhance the feature details of the image during the training process. In the third step, an end-to-end network is established directly through the residual structure for eliminating unnecessary visualization during the process and for ease of training. According to the experimental results, it has been proved that the deep intelligent neural network method proposed here has achieved good results in the application for super-resolution of medical geographic small-target intelligent satellite images.

Based on MicroRNA (miRNA) expression profiles, this article proposes a new algorithm—SVM-RFE-FKNN, which combines the support vector machine-recursive feature elimination (SVM-RFE) algorithm and the fuzzy K-nearest neighbor (FKNN) algorithm, to realize binary classification of tumors. First, the SVM-RFE algorithm was used to select features from the miRNA expression profile dataset to constitute feature subsets and to determine the maximum number of support vectors. Next, this maximum number was regarded as the upper limit of the parameter K in the FKNN algorithm that was then used to classify the samples to be tested. Finally, the leave-one-out cross-validation method was adopted to assess the classification performance of the proposed algorithm. Through experiments, our proposed algorithm was compared with other twelve classification methods, and the result shows that our algorithm had better classification performance. Specifically, with only a few miRNA biomarkers, the proposed algorithm could reach an accuracy of 99.46% and an area under the receiver operating characteristic curve (AUC) of 0.9874.

The stacking algorithm has better generalization ability than other learning algorithms, and can flexibly handle different tasks. The basic model of this algorithm uses heterogeneous learning devices (different types of learning devices), but for each data set in K-fold cross validation, the learners used are homogeneous (the same type of learner). Considering the neglect of the precision difference by a homogeneous heterotopic learner, the accuracy difference weighting method is proposed to improve the traditional stacking algorithm. In the first layer of the traditional stacking algorithm, the algorithm is weighted according to the prediction accuracy, that is, the output of the test set of the first layer is weighted by the weight calculated with the obtained precision, and the weighted result input into the element learner is taken as the feature. As one of the diseases with the highest incidence and mortality, the effective prediction of heart disease can provide an important basis for assisting diagnosis and enhancing the survival rate of patients. In this article, the improved stacking integration algorithm was used to construct a two-layer classifier model to predict heart disease. The experimental results show that the algorithm can effectively improve the prediction accuracy of heart disease through the verification of other heart disease data sets, and it is found that the stacking algorithm has better generalization performance.