OBJECTIVE: To investigate a new noninvasive diagnostic model for nonalcoholic fatty liver disease (NAFLD) based on features of tongue images. METHODS: Healthy controls and volunteers confirmed to have NAFLD by liver ultrasound were recruited from China-Japan Friendship Hospital between September 2018 and May 2019, then the anthropometric indexes and sampled tongue images were measured. The tongue images were labeled by features, based on a brief protocol, without knowing any other clinical data, after a series of corrections and data cleaning. The algorithm was trained on images using labels and several anthropometric indexes for inputs, utilizing machine learning technology. Finally, a logistic regression algorithm and a decision tree model were constructed as 2 diagnostic models for NAFLD. RESULTS: A total of 720 subjects were enrolled in this study, including 432 patients with NAFLD and 288 healthy volunteers. Of them, 482 were randomly allocated into the training set and 238 into the validation set. The diagnostic model based on logistic regression exhibited excellent performance: in validation set, it achieved an accuracy of 86.98%, sensitivity of 91.43%, and specificity of 80.61%; with an area under the curve (AUC) of 0.93 [95% confidence interval (CI) 0.68-0.98]. The decision tree model achieved an accuracy of 81.09%, sensitivity of 91.43%, and specificity of 66.33%; with an AUC of 0.89 (95% CI 0.66-0.92) in validation set. CONCLUSIONS The features of tongue images were associated with NAFLD. Both the 2 diagnostic models, which would be convenient, noninvasive, lightweight, rapid, and inexpensive technical references for early screening, can accurately distinguish NAFLD and are worth further study.
Humans ; Non-alcoholic Fatty Liver Disease/diagnostic imaging* ; Ultrasonography ; Anthropometry ; Algorithms ; China

Humans ; Artificial Intelligence ; Algorithms ; Brain ; Head

OBJECTIVE: To explore an efficient and automatic method for determining the anatomical landmarks of three-dimensional(3D) mandibular data, and to preliminarily evaluate the performance of the method. METHODS: The CT data of 40 patients with normal craniofacial morphology were collected (among them, 30 cases were used to establish the 3D mandibular average model, and 10 cases were used as test datasets to validate the performance of this method in determining the mandibular landmarks), and the 3D mandibular data were reconstructed in Mimics software. Among the 40 cases of mandibular data after the 3D reconstruction, 30 cases that were more similar to the mean value of Chinese mandibular features were selected, and the size of the mandibular data of 30 cases was normalized based on the Procrustes analysis algorithm in MATLAB software. Then, in the Geomagic Wrap software, the 3D mandibular average shape model of the above 30 mandibular data was constructed. Through symmetry processing, curvature sampling, index marking and other processing procedures, a 3D mandible structured template with 18 996 semi-landmarks and 19 indexed mandibular anatomical landmarks were constructed. The open source non-rigid registration algorithm program Meshmonk was used to match the 3D mandible template constructed above with the tested patient's 3D mandible data through non-rigid deformation, and 19 anatomical landmark positions of the patient's 3D mandible data were obtained. The accuracy of the research method was evaluated by comparing the distance error of the landmarks manually marked by stomatological experts with the landmarks marked by the method of this research. RESULTS: The method of this study was applied to the data of 10 patients with normal mandibular morphology. The average distance error of 19 landmarks was 1.42 mm, of which the minimum errors were the apex of the coracoid process [right: (1.01±0.44) mm; left: (0.56±0.14) mm] and maximum errors were the anterior edge of the lowest point of anterior ramus [right: (2.52±0.95) mm; left: (2.57±1.10) mm], the average distance error of the midline landmarks was (1.15±0.60) mm, and the average distance error of the bilateral landmarks was (1.51±0.67) mm. CONCLUSION The automatic determination method of 3D mandibular anatomical landmarks based on 3D mandibular average shape model and non-rigid registration algorithm established in this study can effectively improve the efficiency of automatic labeling of 3D mandibular data features. The automatic determination of anatomical landmarks can basically meet the needs of oral clinical applications, and the labeling effect of deformed mandible data needs to be further tested.
Humans ; Imaging, Three-Dimensional/methods* ; Mandible/diagnostic imaging* ; Software ; Algorithms ; Anatomic Landmarks/anatomy & histology*

Accurate segmentation of retinal blood vessels is of great significance for diagnosing, preventing and detecting eye diseases. In recent years, the U-Net network and its various variants have reached advanced level in the field of medical image segmentation. Most of these networks choose to use simple max pooling to down-sample the intermediate feature layer of the image, which is easy to lose part of the information, so this study proposes a simple and effective new down-sampling method Pixel Fusion-pooling (PF-pooling), which can well fuse the adjacent pixel information of the image. The down-sampling method proposed in this study is a lightweight general module that can be effectively integrated into various network architectures based on convolutional operations. The experimental results on the DRIVE and STARE datasets show that the F₁-score index of the U-Net model using PF-pooling on the STARE dataset improved by 1.98%. The accuracy rate is increased by 0.2%, and the sensitivity is increased by 3.88%. And the generalization of the proposed module is verified by replacing different algorithm models. The results show that PF-pooling has achieved performance improvement in both Dense-UNet and Res-UNet models, and has good universality.
Algorithms ; Retinal Vessels ; Image Processing, Computer-Assisted

OBJECTIVES: Firefighters are prone to suffer from psychological trauma and post-traumatic stress disorder (PTSD) in the workplace, and have a poor prognosis after PTSD. Reliable models for predicting PTSD allow for effective identification and intervention for patients with early PTSD. By collecting the psychological traits, psychological states and work situations of firefighters, this study aims to develop a machine learning algorithm with the aim of effectively and accurately identifying the onset of PTSD in firefighters, as well as detecting some important predictors of PTSD onset. METHODS: This study conducted a cross-sectional survey through convenient sampling of firefighters from 20 fire brigades in Changsha, which were evenly distributed across 6 districts and Changsha County, with a total of 628 firefighters. We used the synthetic minority oversampling technique (SMOTE) to process data sets and used grid search to finish the parameter tuning. The predictive capability of several commonly used machine learning models was compared by 5-fold cross-validation and using the area under the receiver operating characteristic curve (ROC-AUC), accuracy, precision, recall, and F1 score. RESULTS: The random forest model achieved good performance in predicting PTSD with an average AUC score at 0.790. The mean accuracy of the model was 90.1%, with an F1 score of 0.945. The three most important predictors were perseverance, forced thinking, and reflective deep thinking, with weights of 0.165, 0.158, and 0.152, respectively. The next most important predictors were employment time, psychological power, and optimism. CONCLUSIONS PTSD onset prediction model for Changsha firefighters constructed by random forest has strong predictive ability, and both psychological characteristics and work situation can be used as predictors of PTSD onset risk for firefighters. In the next step of the study, validation using other large datasets is needed to ensure that the predictive models can be used in clinical setting.
Humans ; Stress Disorders, Post-Traumatic/diagnosis* ; Firefighters/psychology* ; Cross-Sectional Studies ; Algorithms ; Machine Learning

OBJECTIVE: To explore the application of extended reality (XR) technology in clinical surgeries for improving the success rate of surgeries. METHODS: To assist the surgeons to better understand the location, size and geometric shape of the lesions and reduce potential radiation exposure in minimally invasive surgical navigation based on two-dimensional images, we constructed three-dimensional models based on CT data and used XR technology to achieve intraoperative navigation. An improved quaternion method was used to improve the accuracy of electromagnetic positioning, with which the system error of positioning accuracy was reduced to below 2 mm. A 5G network was used to optimize the server GPU programming algorithm, and real-time video stream coding strategy and network design were adopted to reduce data transmission jam and delay in the remote surgery network, which achieved an average delay of less than 60 ms. A Gaussian distribution deformation model was used to simulate collision detection and stress deformation of the tissues to achieve a tactile perception effect. RESULTS AND CONCLUSION The intraoperative navigation system based on XR technology allowed more accurate determination of the location of the lesions, effectively reduced the surgical risk, and avoided the risk of intraoperative radiation exposure. The low latency and high fidelity of 5G network achieved real-time interaction during the surgery to provide a technical basis for multi-terminal remote cooperative surgery. The combination of force feedback technology and XR technology enables the surgeons to conduct deep immersion preoperative planning and virtual surgery to improve the success rate of surgery and shorten the learning curve.
Algorithms ; Technology

OBJECTIVE: To screen the risk factors for death in patients with nasopharyngeal carcinoma (NPC) using artificial intelligence (AI) technology and establish a risk prediction model. METHODS: The clinical data of NPC patients obtained from SEER database (1973-2015). The patients were randomly divided into model building and verification group at a 7∶3 ratio. Based on the data in the model building group, R software was used to identify the risk factors for death in NPC patients using 4 AI algorithms, namely eXtreme Gradient Boosting (XGBoost), Decision Tree (DT), Least absolute shrinkage and selection operator (LASSO) and random forest (RF), and a risk prediction model was constructed based on the risk factor identified. The C-Index, decision curve analysis (DCA), receiver operating characteristic (ROC) curve and calibration curve (CC) were used for internal validation of the model; the data in the validation group and clinical data of 96 NPC patients (collected from First Affiliated Hospital of Bengbu Medical College) were used for internal and external validation of the model. RESULTS: The clinical data of a total of 2116 NPC patients were included (1484 in model building group and 632 in verification group). Risk factor screening showed that age, race, gender, stage M, stage T, and stage N were all risk factors of death in NPC patients. The risk prediction model for NPC-related death constructed based on these factors had a C-index of 0.76 for internal evaluation, an AUC of 0.74 and a net benefit rate of DCA of 9%-93%. The C-index of the model in internal verification was 0.740 with an AUC of 0.749 and a net benefit rate of DCA of 3%-89%, suggesting a high consistency of the two calibration curves. In external verification, the C-index of this model was 0.943 with a net benefit rate of DCA of 3%-97% and an AUC of 0.851, and the predicted value was consistent with the actual value. CONCLUSIONS Gender, age, race and TNM stage are risk factors of death of NPC patients, and the risk prediction model based on these factors can accurately predict the risks of death in NPC patients.
Humans ; Nasopharyngeal Neoplasms ; Nasopharyngeal Carcinoma ; Artificial Intelligence ; Algorithms ; Software

This study aimed to explore the underlying framework and data characteristics of Tibetan prescription information. The information on Tibetan medicine prescriptions was collected based on 11 Tibetan medicine classics, such as Four Medical Canons(Si Bu Yi Dian). The optimal classification method was used to summarize the information structure of Tibetan medicine prescriptions and sort out the key problems and solutions in data collection, standardization, translation, and analysis. A total of 11 316 prescriptions were collected, involving 139 011 entries and 63 567 pieces of efficacy information of drugs in prescriptions. The information on Tibe-tan medicine prescriptions could be summarized into a "seven-in-one" framework of "serial number-source-name-composition-efficacy-appendix-remarks" and 18 expansion layers, which contained all information related to the inheritance, processing, origin, dosage, semantics, etc. of prescriptions. Based on the framework, this study proposed a "historical timeline" method for mining the origin of prescription inheritance, a "one body and five layers" method for formulating prescription drug specifications, a "link-split-link" method for constructing efficacy information, and an advanced algorithm suitable for the research of Tibetan prescription knowledge discovery. Tibetan medicine prescriptions have obvious characteristics and advantages under the guidance of the theories of "three factors", "five sources", and "Ro-nus-zhu-rjes" of Tibetan medicine. Based on the characteristics of Tibetan medicine prescriptions, this study proposed a multi-level and multi-attribute underlying data architecture, providing new methods and models for the construction of Tibetan medicine prescription information database and knowledge discovery and improving the consistency and interoperability of Tibetan medicine prescription information with standards at all levels, which is expected to realize the "ancient and modern connection-cleaning up the source-data sharing", so as to promote the informatization and modernization research path of Tibetan medicine prescriptions.
Medicine, Tibetan Traditional ; Knowledge Discovery ; Drug Prescriptions ; Databases, Factual ; Algorithms ; Medicine, Chinese Traditional ; Drugs, Chinese Herbal/therapeutic use*

Polysomnography (PSG) monitoring is an important method for clinical diagnosis of diseases such as insomnia, apnea and so on. In order to solve the problem of time-consuming and energy-consuming sleep stage staging of sleep disorder patients using manual frame-by-frame visual judgment PSG, this study proposed a deep learning algorithm model combining convolutional neural networks (CNN) and bidirectional gate recurrent neural networks (Bi GRU). A dynamic sparse self-attention mechanism was designed to solve the problem that gated recurrent neural networks (GRU) is difficult to obtain accurate vector representation of long-distance information. This study collected 143 overnight PSG data of patients from Shanghai Mental Health Center with sleep disorders, which were combined with 153 overnight PSG data of patients from the open-source dataset, and selected 9 electrophysiological channel signals including 6 electroencephalogram (EEG) signal channels, 2 electrooculogram (EOG) signal channels and a single mandibular electromyogram (EMG) signal channel. These data were used for model training, testing and evaluation. After cross validation, the accuracy was (84.0±2.0)%, and Cohen's kappa value was 0.77±0.50. It showed better performance than the Cohen's kappa value of physician score of 0.75±0.11. The experimental results show that the algorithm model in this paper has a high staging effect in different populations and is widely applicable. It is of great significance to assist clinicians in rapid and large-scale PSG sleep automatic staging.
Humans ; Polysomnography ; China ; Sleep Stages ; Sleep ; Algorithms

In this paper, we propose a multi-scale mel domain feature map extraction algorithm to solve the problem that the speech recognition rate of dysarthria is difficult to improve. We used the empirical mode decomposition method to decompose speech signals and extracted Fbank features and their first-order differences for each of the three effective components to construct a new feature map, which could capture details in the frequency domain. Secondly, due to the problems of effective feature loss and high computational complexity in the training process of single channel neural network, we proposed a speech recognition network model in this paper. Finally, training and decoding were performed on the public UA-Speech dataset. The experimental results showed that the accuracy of the speech recognition model of this method reached 92.77%. Therefore, the algorithm proposed in this paper can effectively improve the speech recognition rate of dysarthria.
Humans ; Dysarthria/diagnosis* ; Speech ; Speech Perception ; Algorithms ; Neural Networks, Computer