Quality control system based on artificial intelligence for improving imaging quality of chest CT / 中国医学影像技术

Objective To observe the value of quality control system based on artificial intelligence(AI)for improving imaging quality of chest CT.Methods Totally 1 726 CT images obtained from 415 patients were retrospectively collected,among which 1 414 images were used for convolutional neural network(CNN)training and the rest 312 images were used for validation.Precision,Recall,F1-Score,mean average precision(mAP)and intersection over union(IOU)of quality control system based on AI for chest CT scanning were calculated.Meanwhile,21 patients with unsatisfactory chest CT who would undergo re-examination were prospectively enrolled,and chest CT scanning with quality control system based on AI were performed.The results of 2 examinations were compared.Results Precision,Recall,F1-Score,mAP and IOU of quality control system based on AI for chest CT were all good.All 21 cases were diagnosed correctly with re-examination CT based on quality control system.Among 21 cases,the first CT misdiagnosed 19 cases,the displaying of the area,volume and display quality of pulmonary nodules were not significantly different,but the morphology,boundaries,spiny protrusions,vacuolar signs,inflatable bronchial signs of nodules as well as the thickened and twisted blood vessels were obviously different between 2 times examination.The first CT missed 1 case while correctly diagnosed 1 case.Conclusion The quality control system based on AI was helpful for improving imaging quality of chest CT and increasing diagnostic efficacy.

Deep neural networks analysis of 18F-FDG PET imaging in postoperative patients with temporal lobe epilepsy / 中华核医学与分子影像杂志

Objective:To predict the short-term postoperative recurrence status of patients with refractory temporal lobe epilepsy (TLE) by analyzing preoperative 18F-FDG PET images and patients′ clinical characteristics based on deep residual neural network (ResNet). Methods:Retrospective analysis was conducted on preoperative 18F-FDG PET images and clinical data of 220 patients with refractory TLE (132 males and 88 females, age 23.0(20.0, 30.2) years)) in the First Affiliated Hospital of Jinan University between January 2014 and June 2020. ResNet was used to perform high-throughput feature extraction on preprocessed PET images and clinical features, and to perform a postoperative recurrence prediction task for differentiating patients with TLE. The predictive performance of ResNet model was evaluated by ROC curve analysis, and the AUC was compared with that of classical Cox proportional risk model using Delong test. Results:Based on PET images combined with clinical feature training, AUCs of the ResNet in predicting 12-, 24-, and 36-month postoperative recurrence were 0.895±0.073, 0.861±0.058 and 0.754±0.111, respectively, which were 0.717±0.093, 0.697±0.081 and 0.645±0.087 for Cox proportional hazards model respectively ( z values: -3.00, -2.98, -1.09, P values: 0.011, 0.018, 0.310). The ResNet showed best predictive effect for recurrence events within 12 months after surgery. Conclusion:The ResNet model is expected to be used in clinical practice for postoperative follow-up of patients with TLE, helping for risk stratification and individualized management of postoperative patients.

Machine learning in predicting severe acute respiratory infection outbreaks / Aprendizado de máquina na previsão de surtos de síndrome respiratória aguda grave / Aprendizaje automático en la predicción de brotes de síndrome respiratorio agudo grave

Abstract: Severe acute respiratory infection (SARI) outbreaks occur annually, with seasonal peaks varying among geographic regions. Case notification is important to prepare healthcare networks for patient attendance and hospitalization. Thus, health managers need adequate resource planning tools for SARI seasons. This study aims to predict SARI outbreaks based on models generated with machine learning using SARI hospitalization notification data. In this study, data from the reporting of SARI hospitalization cases in Brazil from 2013 to 2020 were used, excluding SARI cases caused by COVID-19. These data were prepared to feed a neural network configured to generate predictive models for time series. The neural network was implemented with a pipeline tool. Models were generated for the five Brazilian regions and validated for different years of SARI outbreaks. By using neural networks, it was possible to generate predictive models for SARI peaks, volume of cases per season, and for the beginning of the pre-epidemic period, with good weekly incidence correlation (R2 = 0.97; 95%CI: 0.95-0.98, for the 2019 season in the Southeastern Brazil). The predictive models achieved a good prediction of the volume of reported cases of SARI; accordingly, 9,936 cases were observed in 2019 in Southern Brazil, and the prediction made by the models showed a median of 9,405 (95%CI: 9,105-9,738). The identification of the period of occurrence of a SARI outbreak is possible using predictive models generated with neural networks and algorithms that employ time series.

Resumo: Surtos de síndrome respiratória aguda grave (SRAG) ocorrem anualmente, com picos sazonais variando entre regiões geográficas. A notificação dos casos é importante para preparar as redes de atenção à saúde para o atendimento e internação dos pacientes. Portanto, os gestores de saúde precisam ter ferramentas adequadas de planejamento de recursos para as temporadas de SRAG. Este estudo tem como objetivo prever surtos de SRAG com base em modelos gerados com aprendizado de máquina usando dados de internação por SRAG. Foram incluídos dados sobre casos de hospitalização por SRAG no Brasil de 2013 a 2020, excluindo os casos causados pela COVID-19. Estes dados foram preparados para alimentar uma rede neural configurada para gerar modelos preditivos para séries temporais. A rede neural foi implementada com uma ferramenta de pipeline. Os modelos foram gerados para as cinco regiões brasileiras e validados para diferentes anos de surtos de SRAG. Com o uso de redes neurais, foi possível gerar modelos preditivos para picos de SRAG, volume de casos por temporada e para o início do período pré-epidêmico, com boa correlação de incidência semanal (R2 = 0,97; IC95%: 0,95-0,98, para a temporada de 2019 na Região Sudeste). Os modelos preditivos obtiveram uma boa previsão do volume de casos notificados de SRAG; dessa forma, foram observados 9.936 casos em 2019 na Região Sul, e a previsão feita pelos modelos mostrou uma mediana de 9.405 (IC95%: 9.105-9.738). A identificação do período de ocorrência de um surto de SRAG é possível por meio de modelos preditivos gerados com o uso de redes neurais e algoritmos que aplicam séries temporais.

Resumen: Brotes de síndrome respiratorio agudo grave (SRAG) ocurren todos los años, con picos estacionales que varían entre regiones geográficas. La notificación de los casos es importante para preparar las redes de atención a la salud para el cuidado y hospitalización de los pacientes. Por lo tanto, los gestores de salud deben tener herramientas adecuadas de planificación de recursos para las temporadas de SRAG. Este estudio tiene el objetivo de predecir brotes de SRAG con base en modelos generados con aprendizaje automático utilizando datos de hospitalización por SRAG. Se incluyeron datos sobre casos de hospitalización por SRAG en Brasil desde 2013 hasta 2020, salvo los casos causados por la COVID-19. Se prepararon estos datos para alimentar una red neural configurada para generar modelos predictivos para series temporales. Se implementó la red neural con una herramienta de canalización. Se generaron los modelos para las cinco regiones brasileñas y se validaron para diferentes años de brotes de SRAG. Con el uso de redes neurales, se pudo generar modelos predictivos para los picos de SRAG, el volumen de casos por temporada y para el inicio del periodo pre-epidémico, con una buena correlación de incidencia semanal (R2 = 0,97; IC95%: 0,95-0,98, para la temporada de 2019 en la Región Sudeste). Los modelos predictivos tuvieron una buena predicción del volumen de casos notificados de SRAG; así, se observaron 9.936 casos en 2019 en la Región Sur, y la predicción de los modelos mostró una mediana de 9.405 (IC95%: 9.105-9.738). La identificación del periodo de ocurrencia de un brote de SRAG es posible a través de modelos predictivos generados con el uso de redes neurales y algoritmos que aplican series temporales.

Predictive model for urosepsis in patients with Upper Urinary Tract Calculi based on ultrasonography and urinalysis using artificial intelligence learning

ABSTRACT Purpose To construct a predicting model for urosepsis risk for patients with upper urinary tract calculi based on ultrasound and urinalysis. Materials and Methods A retrospective study was conducted in patients with upper urinary tract calculi admitted between January 2016 and January 2020. The patients were randomly grouped into the training and validation sets. The training set was used to identify the urosepsis risk factors and construct a risk prediction model based on ultrasound and urinalysis. The validation set was used to test the performance of the artificial neural network (ANN). Results Ultimately, 1716 patients (10.8% cases and 89.2% control) were included. Eight variables were selected for the model: sex, age, body temperature, diabetes history, urine leukocytes, urine nitrite, urine glucose, and degree of hydronephrosis. The area under the receiver operating curve in the validation and training sets was 0.945 (95% CI: 0.903-0.988) and 0.992 (95% CI: 0.988-0.997), respectively. Sensitivity, specificity, and Yuden index of the validation set (training set) were 80.4% (85.9%), 98.2% (99.0%), and 0.786 (0.849), respectively. Conclusions A preliminary screening model for urosepsis based on ultrasound and urinalysis was constructed using ANN. The model could provide risk assessments for urosepsis in patients with upper urinary tract calculi.

The development and primary application of a deep learning convolutional neural network in the field of revision total hip arthroplasty CT segmentation / 中华骨科杂志

Objective:To develop a preoperative CT image segmentation algorithm based on artificial intelligence deep learning technology for total hip arthroplasty (THA) revision surgery, and to verify and preliminarily apply it.Methods:A total of 706 revision cases with clear CT data from April 2019 to October 2022 in Chinese PLA General Hospital were retrospectively analyzed, including 520 males, aged 58.45±18.13 years, and 186 females, aged 52.23±16.23 years. All of them were unilateral, and there were 402 hips on the left and 304 hips on the right. The transformer_unet convolutional neural network was constructed and trained using Tensorflow 1.15 to achieve intelligent segmentation of the revision THA CT images. Based on the developed three-dimensional planning system of total hip arthroplasty, an intelligent planning system for revision hip arthroplasty was preliminarily constructed. Dice overlap coefficient (DOC), average surface distance (ASD) and Hausdorff distance (HD) parameters were used to evaluate the segmentation accuracy of transformer_unet, full convolution network (FCN), 2D U-shaped Net and Deeplab v3 +, and segmentation time was used to evaluate the segmentation efficiency of these networks.Results:Compared with the FCN, 2D U-Net, and Deeplab v3+ learning curves, the transformer_unet network could achieve better training effect with less training amount.The DOC of transformer_unet was 95%±4%, the HD was 3.35±1.03 mm, and the ASD was 1.38±0.02 mm; FCN was 94%±4%, 4.83±1.90 mm, 1.42±0.03 mm; 2D U-Net was 93%±5%, 5.27±2.20 mm, and 1.46±0.02 mm, respectively. Deeplab v3+ was 92%±4%, 6.12±1.84 mm, 1.52±0.03 mm, respectively. The transformer_unet coefficients were better than those of the other three convolutional neural networks, and the differences were statistically significant (all P<0.05). The segmentation time of transformer_unet was 0.031±0.001 s, FCN was 0.038±0.002 s, 2D U-Net was 0.042±0.001 s, Deeplab v3+ was 0.048±0.002 s. The segmentation time of transformer_unet was less than that of the other three convolutional neural networks, and the difference was statistically significant ( P<0.05). Based on the results of previous studies, an artificial intelligence assisted preoperative planning system for THA revision surgery was initially constructed. Conclusion:Compared with FCN, 2D U-Net and Deeplab v3+, the transformer_unet convolutional neural network can complete the segmentation of the revision THA CT image more accurately and efficiently, which is expected to provide technical support for preoperative planning and surgical robots.

Value of machine learning model based on 18F-FDG PET/CT radiomics features in differential diagnosis between gastric cancer and primary gastric lymphoma / 中华核医学与分子影像杂志

Objective:To investigate the value of machine learning model based on 18F-FDG PET/CT radiomics features in preoperative differential diagnosis of gastric cancer (GC) and primary gastric lymphoma (PGL). Methods:A total of 155 patients with GC (104 males, 51 females; age (59.3±12.8) years) and 82 patients with PGL (40 males, 42 females; age (56.8±14.6) years) who underwent 18F-FDG PET/CT imaging before treatment from January 2012 to December 2020 in Tianjin Medical University Cancer Institute and Hospital were included in this retrospective study. Patients were randomly divided into training set and test set by using Python3.7.1 software. Volumes of interest (VOIs) in PET and CT images were drawn and two-dimensional and three-dimensional radiomics features were extracted. Two machine learning models, including multi-layer perceptron (MLP) and support vector machine (SVM), were established based on CT radiomics features alone, PET radiomics features alone and PET/CT radiomics features to differentiate GC and PGL, respectively. The predictive performance of each model was evaluated by ROC curve analysis. Results:There were 166 patients in training set and 71 patients in test set. Generally, SVM machine learning model based on PET/CT radiomics features showed a trend to be superior to MLP machine learning model in the differential diagnosis of GC and PGL (PET-SVM: AUC=0.88, 95% CI: 0.83-0.94); PET/CT-MLP: AUC=0.80, 95% CI: 0.73-0.87; z=1.15, P=0.337). The AUC of PET/CT-SVM machine learning model was significantly higher than that of CT-SVM machine learning model (CT-SVM: AUC=0.74, 95% CI: 0.67-0.81; z=2.28, P=0.022). Conclusion:Machine learning model based on 18F-FDG PET/CT radiomics features is expected to be a non-invasive, effective tool for preoperative differential diagnosis of GC and PGL.

Application analysis of digital neural network-based data mining method in maximizing the performance of sports training / Análise de aplicação do método de mineração de dados baseado em rede neural digital na maximização do desempenho do treinamento esportivo / Análisis de la aplicación del método de minería de datos basado en redes neuronales digitales para maximizar el rendimiento del entrenamiento deportivo

ABSTRACT Introduction: In today's rapid development of science and technology, digital network data mining technology is developing as fast as the expansion of the frontiers of science and technology allows, with a very broad application level, covering most of the civilized environment. However, there is still much to explore in the application of sports training. Objective: Analyze the feasibility of data mining based on the digital network of sports training, maximizing athletes' training. Methods: This paper uses the experimental analysis of human FFT, combined with BP artificial intelligence network and deep data mining technology, to design a new sports training environment. The controlled test of this model was designed to compare advanced athletic training modalities with traditional modalities, comparing the athletes' explosive power, endurance, and fitness. Results: After 30 days of physical training, the athletic strength of athletes with advanced fitness increased by 15.33%, endurance increased by 15.85%, and fitness increased by 14.23%. Conclusion: The algorithm designed in this paper positively impacts maximizing athletes' training. It may have a favorable impact on training outcomes, as well as increase the athlete's interest in the sport. Level of evidence II; Therapeutic studies - investigating treatment outcomes.

RESUMO Introdução: No rápido desenvolvimento atual de ciência e tecnologia, a tecnologia de mineração de dados de rede digital desenvolve-se tão rápido quanto a expansão das fronteiras da ciência e tecnologia permitem, com um nível de aplicação muito amplo, cobrindo a maior parte do ambiente civilizado. No entanto, ainda há muito para explorar da aplicação no treinamento esportivo. Objetivo: Análise de viabilidade da mineração de dados com base na rede digital da formação esportiva, maximizar o treinamento dos atletas. Métodos: Este trabalho utiliza a análise experimental da FFT humana, combinada com a rede de inteligência artificial da BP e tecnologia de mineração profunda de dados, para projetar um novo ambiente de treinamento esportivo. O teste controlado deste modelo foi projetado para comparar modalidades avançadas de treinamento atlético com as modalidades tradicionais, comparando o poder explosivo, resistência e condição física do atleta. Resultados: Após 30 dias de treinamento físico, a força atlética dos esportistas com aptidão física avançada aumentou 15,33%, a resistência aumentou 15,85%, e o condicionamento físico aumentou 14,23%. Conclusão: O algoritmo desenhado neste artigo tem um impacto positivo na maximização do treinamento dos atletas. Pode ter um impacto favorável nos resultados do treinamento, bem como aumentar o interesse do atleta pelo esporte. Nível de evidência II; Estudos terapêuticos - investigação dos resultados do tratamento.

RESUMEN Introducción: En el rápido desarrollo actual de la ciencia y la tecnología, la tecnología de extracción de datos de redes digitales se desarrolla tan rápido como lo permiten las fronteras en expansión de la ciencia y la tecnología, con un nivel de aplicación muy amplio que abarca la mayor parte del entorno civilizado. Sin embargo, aún queda mucho por explorar de la aplicación en el entrenamiento deportivo. Objetivo: Análisis de viabilidad de la minería de datos basada en la red digital de entrenamiento deportivo, maximizar la formación de los atletas. Métodos: Este trabajo utiliza el análisis experimental de la FFT humana, combinado con la red de inteligencia artificial BP y la tecnología de minería de datos profunda, para diseñar un nuevo entorno de entrenamiento deportivo. La prueba controlada de este modelo se diseñó para comparar las modalidades de entrenamiento atlético avanzado con las modalidades tradicionales, comparando la potencia explosiva, la resistencia y la forma física del atleta. Resultados: Después de 30 días de entrenamiento físico, la fuerza atlética de los atletas con un estado físico avanzado aumentó en un 15,33%, la resistencia aumentó en un 15,85% y el estado físico aumentó en un 14,23%. Conclusión: El algoritmo diseñado en este trabajo tiene un impacto positivo en la maximización del entrenamiento de los atletas. Puede tener un impacto favorable en los resultados del entrenamiento, así como aumentar el interés del atleta por el deporte. Nivel de evidencia II; Estudios terapéuticos - investigación de los resultados del tratamiento.

Automatic diagnosis and assessment of bone metastases on bone scans based on deep learning / 中华核医学与分子影像杂志

Objective:To develop an approach for the automatic diagnosis of bone metastasis and to design a parameter of quantitative evaluation for tumor burden on bone scans based on deep learning technology.Methods:A total of 621 cases (389 males, 232 females, age: 12-93 years) of bone scan images from the Department of Nuclear Medicine in Tenth People′s Hospital of Tongji University from March 2018 to July 2019 were retrospectively analyzed. Images were divided into bone metastasis group and non-bone metastasis group. Eighty percent of the cases were randomly extracted from both groups as the training set, and the rest of cases were used as the test set. A deep residual convolutional neural network ResNet34 was used to construct the classification model and the segmentation model. The sensitivity, specificity and accuracy were calculated and the performance differences of the classification model in different age groups (15 cases of <50 years, 75 cases of ≥50 and <70 years, 33 cases of ≥70 years) were analyzed. The regions of metastatic bone lesions were automatically segmented by the segmentation model. The Dice coefficient was used to evaluate the effect of the segmentation model and the manual labeled results. Finally, the bone scans tumor burden index (BSTBI) was calculated to assess the tumor burden of bone metastases.Results:There were 280 cases with bone metastases and 341 cases with non-bone metastases, including 498 in training set and 123 in test set. The classification model could accurately identify bone metastases, with the sensitivity, specificity and accuracy of 92.59%(50/54), 85.51%(59/69) and 88.62%(109/123), respectively, and it performed best in the <50 years group (sensitivity, 2/2; specificity, 12/13; accuracy, 14/15). The specificity in the ≥70 years group (8/12) was the lowest. The Dice coefficient of bone metastatic area and bladder area were 0.739 and 0.925 in the segmentation model, which performed similarly in the three age groups. Preliminary results showed that the value of BSTBI increased with the increase of the number of bone metastatic lesions and the degree of 99Tc m-MDP uptake. The machine learning model in this study took (0.48±0.07) s for the entire analysis process from input to the final BSTBI calculation. Conclusions:The deep learning based on automatic diagnosis framework for bone metastases can automatically and accurately identify segment bone metastases and calculate tumor burden. It provides a new way for the interpretation of bone scans. The proposed BSTBI may be used as a quantitative evaluation indicator in the future to assess the tumor burden of bone metastases based on bone scans.

Study on improving the quality of low-dose PET images of children based on generative adversarial networks / 中华核医学与分子影像杂志

Objective:To investigate the value of generative adversarial networks-based PET image reconstruction in improving the quality of low-dose 18F-FDG PET images and lesion detection in pediatric patients. Methods:Retrospective analysis of 61 PET images of children (38 males, 23 females, age (4.0±3.5) years) who underwent 18F-FDG total-body PET/CT imaging in Beijing Friendship Hospital, Capital Medical University from August 2021 to December 2021 was performed. The low-dose images (30 s, 20 s, 10 s) of all children extracted by list mode were input into the generative adversarial networks for deep learning (DL) reconstruction to obtain the corresponding simulated standard full-dose images (DL-30 s, DL-20 s, DL-10 s). The semi-quantitative parameters of the liver blood pool and primary lesion of standard full-dose 120 s, 30 s, 20 s, 10 s, DL-30 s, DL-20 s, and DL-10 s images were measured. The target-to-background ratio (TBR), contrast-to-noise ratio (CNR), and CV were calculated. The 5-point Likert scale was used for subjective scoring of image quality, and the detective abilities for positive lesions of each groups were compared. The sensitivities and positive predictive values of positive lesions detection were calculated. Mann-Whitney U test and Kruskal-Wallis rank sum test and χ2 test were used for data analyses. Results:CNR of the 30 s, 20 s, and 10 s groups were lower than those of DL-30 s, DL-20 s, and DL-10 s groups, respectively ( z values: -3.58, -3.20, -3.65, all P<0.05). Score of DL-10 s group was significantly lower than those of 120 s, DL-30 s and DL-20 s groups (4(3, 4), 5(4, 5), 4(4, 5), 4(4, 5); H=97.70, P<0.001). There were no significant differences in TBR, CNR, CV, SUV max and SUV mean of lesions and liver blood pool in 120 s, DL-30 s, DL-20 s, and DL-10 s groups ( H values: 0.00-6.76, all P>0.05). The sensitivities of positive lesion detection in DL-30 s, DL-20 s, and DL-10 s groups were 97.83%(225/230), 96.96%(223/230), 95.65%(220/230), respectively, and the positive predictive values were 96.57%(225/233), 93.70%(223/238), 84.94%(220/259), respectively. The positive predictive value in DL-10 s group was lower than those in DL-30 s and DL-20 s groups ( χ2=23.51, P<0.001). There were more false-positive and false-negative lesions detected by DL-10 s group than those of DL-30 s and DL-20 s groups in different sites. Conclusion:Based on the generative adversarial networks, the image quality of DL-20 s group is high and can meet the clinical diagnostic requirements.

COPD identification using maximum intensity projection of lung field CT images and deep convolution neural network / 中华健康管理学杂志

Objective:To propose a model using the maximum intensity projection (MIP) of lung field computed tomography (CT) images and deep convolution neural network (CNN) and explore its value in identifying chronic obstructive pulmonary disease (COPD).Methods:A total of 201 subjects were selected from the Second Hospital of Dalian Medical University from January 2010 to May 2021. All subjects were included according to the inclusion criteria and were divided into COPD group (101 cases) and healthy controls group (100 cases). Each patient underwent a high-resolution CT scan of the chest and pulmonary function test. First, the lung field was extracted from CT images and the intrapulmonary MIP images were acquired. Second, with these MIP images as input, the model for identifying COPD was constructed based on a modified residual network (ResNet). Finally, the influence of the number of residual blocks on the performance of the models was investigated. Accuracy, sensitivity, specificity, positive predictive value, negative predictive value, receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to evaluate the identification efficiency.Results:The accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) of ResNet26 was 76.1%, 76.2%, 76.0%, 76.2%, and 76.0%, respectively; and the AUC of the test was 0.855 (95% CI: 0.799-0.901). The accuracy, sensitivity, specificity, PPV, NPV of ResNet50 was 77.6%, 76.2%, 79.0%, 78.6%, and 76.7%, respectively; and the AUC of the test was 0.854 (95% CI: 0.797-0.900). The accuracy, sensitivity, specificity, PPV, NPV of ResNet26d was 82.1%, 83.2%, 81.0%, 81.6%, and 82.7%, respectively; and the AUC of the test was 0.885 (95% CI: 0.830-0.926). Conclusions:The COPD identification model via MIP images from CT images within the lung and deep CNN is successfully constructed and achieves accurate COPD identification. And it can provide an effective tool for COPD screening.

Development and validation of multilayer perceptual neural network in glomerular filtration rate evaluation / 中华肾脏病杂志

Objective:To develop a neural network model for the evaluation of glomerular filtration rate (GFR) based on multilayer perceptual neural network, and to compare with the improved Chinese based creatinine GFR evaluation formula (C-GFR cr) and the evaluation formula (EPI-GFR cr) of the American Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) for the clinical applicability of multilayer perceptual neural network model in evaluating GFR. Methods:A total of 684 chronic kidney disease (CKD) patients used for developing a modified version of China′s based creatinine GFR evaluation formula were taken as the research object. The data of 454 patients were randomly selected as the development group and the data of the other 230 patients were as the verification group. The multilayer perceptual neural network GFR evaluation model (M-GFR cr) was established. With the double plasma GFR as the reference value (rGFR), the correlation, mean difference, mean absolute difference, precision and accuracy of C-GFR cr, EPI-GFR cr and M-GFR cr were compared. Results:Among the 684 CKD patients, there were 352 males and 332 females, with age of (49.9±15.8) years. The correlation between M-GFR cr and rGFR was the highest (Pearson correlation =0.93, P<0.001). The mean difference of M-GFR cr was lower than that of C-GFR cr ( Z=9.929, P<0.001) and EPI-GFR cr ( Z=10.573, P<0.001). The mean absolute difference of M-GFR cr was also lower than that of C-GFR cr ( Z=3.953, P<0.001) and EPI-GFR cr ( Z=4.210, P<0.001). The accuracy of ±15% of M-GFR cr was higher than that of C-GFR cr ( χ2=26.068, P<0.001) and EPI-GFR cr ( χ2=23.154, P<0.001). The accuracy of ±30% of M-GFR cr was also higher than that of C-GFR cr ( χ2=8.264, P=0.001) and EPI-GFR cr ( χ2=11.963, P=0.001). The results of different stages of CKD showed that in the early stage of CKD (CKD 1-2), the mean difference of M-GFR cr was lower than that of C-GFR cr ( Z=7.401, P<0.001) and EPI-GFR cr ( Z=8.096, P<0.001); the mean absolute difference of M-GFR cr was also lower than that of C-GFR cr ( Z=4.723, P<0.001) and EPI-GFR cr ( Z=4.946, P<0.001); the accuracy of ±15% of M-GFR cr was higher than that of C-GFR cr ( χ2=23.547, P<0.001) and EPI-GFR cr ( χ2=26.421, P<0.001); the accuracy of ±30% of M-GFR cr was also higher than that of C-GFR cr ( χ2=12.089, P=0.001) and EPI-GFR cr ( χ2=16.168, P<0.001). But there was no significant difference in the applicability among C-GFR cr, EPI-GFR cr and M-GFR cr in the advanced stages of CKD (CKD 3-5). Conclusion:Compared with the improved Chinese based creatinine GFR evaluation formula C-GFR cr and CKD-EPI evaluation formula EPI-GFR cr, the accuracy of multilayer perceptual neural network model to evaluate GFR in CKD patients has been significantly improved, especially in CKD 1-2 stage.

Fundus tessellation segmentation and quantization based on the deep convolution neural network / 中华眼底病杂志

Objective:To propose automatic measurement of global and local tessellation density on color fundus images based on a deep convolutional neural network (DCNN) method.Methods:An applied study. An artificial intelligence (AI) database was constructed, which contained 1 005 color fundus images captured from 1 024 eyes of 514 myopic patients in the Northern Hospital of Qingdao Eye Hospital from May to July, 2021. The images were preprocessed by using RGB color channel re-calibration method (CCR algorithm), CLAHE algorithm based on Lab color space, Retinex algorithm for multiple iterative illumination estimation, and multi-scale Retinex algorithm. The effects on the segmentation of tessellation by adopting the abovemetioned image enhancement methods and utilizing the Dice, Edge Overlap Rate and clDice loss were compared and observed. The tessellation segmentation model for extracting the tessellated region in the full fundus image as well as the tissue detection model for locating the optic disc and macular fovea were built up. Then, the fundus tessellation density (FTD), macular tessellation density (MTD) and peripapillary tessellation density (PTD) were calculated automatically.Results:When applying CCR algorithm for image preprocessing and the training losses combination strategy, the Dice coefficient, accuracy, sensitivity, specificity and Jordan index for fundus tessellation segmentation were 0.723 4, 94.25%, 74.03%, 96.00% and 70.03%, respectively. Compared with the manual annotations, the mean absolute errors and root mean square errors of FTD, MTD, PTD automatically measured by the model were 0.014 3, 0.020 7, 0.026 7 and 0.017 8, 0.032 3, 0.036 5, respectively.Conclusion:The DCNN-based segmentation and detection method can automatically measure the tessellation density in the global and local regions of the fundus of myopia patients, which can more accurately assist clinical monitoring and evaluation of the impact of fundus tessellation changes on the development of myopia.

Multimodal deep learning model for staging diabetic retinopathy based on ultra-widefield fluorescence angiography / 中华眼底病杂志

Objective:To apply the multi-modal deep learning model to automatically classify the ultra-widefield fluorescein angiography (UWFA) images of diabetic retinopathy (DR).Methods:A retrospective study. From 2015 to 2020, 798 images of 297 DR patients with 399 eyes who were admitted to Eye Center of Renmin Hospital of Wuhan University and were examined by UWFA were used as the training set and test set of the model. Among them, 119, 171, and 109 eyes had no retinopathy, non-proliferative DR (NPDR), and proliferative DR (PDR), respectively. Localization and assessment of fluorescein leakage and non-perfusion regions in early and late orthotopic images of UWFA in DR-affected eyes by jointly optimizing CycleGAN and a convolutional neural network (CNN) classifier, an image-level supervised deep learning model. The abnormal images with lesions were converted into normal images with lesions removed using the improved CycleGAN, and the difference images containing the lesion areas were obtained; the difference images were classified by the CNN classifier to obtain the prediction results. A five-fold cross-test was used to evaluate the classification accuracy of the model. Quantitative analysis of the marker area displayed by the differential images was performed to observe the correlation between the ischemia index and leakage index and the severity of DR.Results:The generated fake normal image basically removed all the lesion areas while retaining the normal vascular structure; the difference images intuitively revealed the distribution of biomarkers; the heat icon showed the leakage area, and the location was basically the same as the lesion area in the original image. The results of the five-fold cross-check showed that the average classification accuracy of the model was 0.983. Further quantitative analysis of the marker area showed that the ischemia index and leakage index were significantly positively correlated with the severity of DR ( β=6.088, 10.850; P<0.001). Conclusion:The constructed multimodal joint optimization model can accurately classify NPDR and PDR and precisely locate potential biomarkers.

Experimental research on forecasting index of biological image aerobic exercise analysis of artificial neural network / Pesquisa experimental sobre índice de previsão de imagem biológica análise de exercício aeróbio de rede neural artificial / Investigación experimental sobre índice de pronóstico de imágenes biológicas análisis de ejercicio aeróbico de redes neuronales artificiales

ABSTRACT Objective: The paper uses artificial neural network images to explore the effects of aerobic exercise on the gamma rhythm of theta period in the awake hippocampal CA1 area of APP/PS1/tau mice and the low-frequency gamma rhythm of the sleep state hippocampal CA1 area SWR period. Methods: Clean grade 6-month-old APP/PS1/tau mice were randomly divided into quiet group (AS) and exercise group (AE), C57BL/6J control group mice were randomly divided into quiet group (CS) and exercise group (CE). The AE group and the CE group performed 12-week treadmill exercise, 5d/week, 60min/d, the first 10min exercise load was 12m/min, the last 50min was 15m/min treadmill slope was 0°. Eight-arm maze detection of behavioral changes in mice; multi-channel in vivo recording technology to record the electrical signals of the awake state and sleep state in the hippocampal CA1 area, MATLAB extracts the awake state theta period and sleep state SWR period, multi-window spectrum estimation method Perform time-frequency analysis and power spectral density analysis. Results: 12 weeks of aerobic exercise can significantly improve the working memory and reference memory of the AS group, increase the gamma energy in theta period of the awake hippocampus CA1 area and the low-frequency gamma energy in the sleep state CA1 area SWR period. Conclusions: Aerobic exercise can improve the neural network state of the AD model and increase the gamma energy in theta period of the hippocampus CA1 area, and the low-frequency gamma energy in the SWR period is one of the neural network mechanisms for its overall behavioral improvement. Level of evidence II; Therapeutic studies - investigation of treatment results.

RESUMO Objetivo: o artigo usa imagens de redes neurais artificiais para explorar os efeitos do exercício aeróbio no ritmo gama do período teta na área CA1 do hipocampo desperto de camundongos APP/PS1/tau e o ritmo gama de baixa frequência da área CA1 do hipocampo do estado de sono Período SWR. Métodos: Camundongos APP/PS1/tau de grau limpo de 6 meses de idade foram divididos aleatoriamente em grupo quieto (AS) e grupo de exercício (AE), os camundongos do grupo controle C57BL/6J foram divididos aleatoriamente em grupo quieto (CS) e grupo de exercício (CE). O grupo AE e o grupo CE realizaram 12 semanas de exercício em esteira, 5d/semana, 60min/d, a primeira carga de exercício de 10min foi de 12m/min, a última de 50min foi de 15m/min e a inclinação da esteira foi de 0 °. Detecção de labirinto de oito braços de mudanças comportamentais em camundongos; tecnologia de gravação in vivo multicanal para registrar os sinais elétricos do estado de vigília e do estado de sono na área CA1 do hipocampo, MATLAB extrai o período de tempo teta do estado de vigília e o período de tempo SWR do estado de sono, método de estimativa de espectro de múltiplas janelas. e análise de densidade espectral de potência. Resultados: 12 semanas de exercícios aeróbicos podem melhorar significativamente a memória de trabalho e a memória de referência do grupo AS, aumentar a energia gama no período teta da área CA1 do hipocampo acordado e a energia gama de baixa frequência na área CA1 do estado de sono período SWR. Conclusões: O exercício aeróbico pode melhorar o estado da rede neural do modelo AD e aumentar a energia gama no período teta da área CA1 do hipocampo e a energia gama de baixa frequência no período SWR é um dos mecanismos da rede neural para seu comportamento geral. Nível de evidência II; Estudos terapêuticos- investigação dos resultados do tratamento.

RESUMEN Objetivo: El artículo utiliza imágenes de redes neuronales artificiales para explorar los efectos del ejercicio aeróbico en el ritmo gamma del período theta en el área CA1 del hipocampo despierto de ratones APP/PS1/tau y el ritmo gamma de baja frecuencia del área CA1 del hipocampo en estado de sueño. Período de ROE. Métodos: Se dividieron aleatoriamente ratones APP/PS1/tau de 6 meses de edad de grado limpio en grupo tranquilo (AS) y grupo de ejercicio (AE), los ratones del grupo de control C57BL/6J se dividieron aleatoriamente en grupo tranquilo (CS) y grupo de ejercicio (CE). El grupo de EA y el grupo de EC realizaron 12 semanas de ejercicio en cinta rodante, 5 días a la semana, 60 min/d, la primera carga de ejercicio de 10 min fue de 12 m/min, los últimos 50 min fueron de 15 m/min y la pendiente de la cinta fue de 0 °. Detección en laberinto de ocho brazos de cambios de comportamiento en ratones; tecnología de grabación in vivo multicanal para registrar las señales eléctricas del estado despierto y del estado de sueño en el área CA1 del hipocampo, MATLAB extrae el período de tiempo theta del estado despierto y el período de tiempo de SWR del estado de suspensión, método de estimación de espectro de múltiples ventanas Realizar análisis de tiempo-frecuencia y análisis de densidad espectral de potencia. Resultados: 12 semanas de ejercicio aeróbico pueden mejorar significativamente la memoria de trabajo y la memoria de referencia del grupo AS, aumentar la energía gamma en el período theta del área CA1 del hipocampo despierto y la energía gamma de baja frecuencia en el período SWR del área CA1 del estado de sueño. Conclusiones: El ejercicio aeróbico puede mejorar el estado de la red neuronal del modelo AD y aumentar la energía gamma en el período theta del área del hipocampo CA1 y la energía gamma de baja frecuencia en el período SWR es uno de los mecanismos de la red neuronal para su comportamiento general. Nivel de evidencia II; Estudios terapéuticos- investigación de los resultados del tratamiento.

Analysis of biolo artificial neural network in prediction of aerobic exercise index based on algorithm / Análise de rede neural artificial na predição do índice de exercício aeróbio baseada em algorítimo / Análisis de red neuronal artificial en la predicción del índice de ejercicio aeróbico

ABSTRACT Objective: To study the relationship between aerobic activity and cardiac autonomic nerve activity by artificial neural network algorithm and biological image fusion; because of the artificial neural network model (ANN) problems, biological image processing technology is introduced based on ANN. Methods: An Ann under biological image intelligence algorithm is proposed, a classifier suitable for electrocardiograph (ECG) screening is designed, and an ECG signal screening system is successfully established. Moreover, the data set of normal recovered ECG signals of the subjects during the experimental period is constructed, and a classifier is used to extract the characteristic data of a normal ECG signal during the experimental period. Results: The changes in resting heart rate and other physical health indicators are analyzed by combining resting physiological indicators, namely heart rate, body weight, body mass index and body fat rate. The results show that the self-designed classifier can efficiently process the ECG images, and long-term regular activities can improve the physical conditions of most people. Most subjects' body weight and body fat rate decrease with the extension of experiment time, and the resting heart rate decreases relatively. Conclusions: Certain indicators can be used to predict a person's dynamic physical health, which indicates that the experimental research of index prediction in this research has a good effect, which not only extends the application of artificial neural network but also lays a foundation for the research and implementation of ECG intelligent testing wearable devices. Level of evidence II; Therapeutic studies - investigation of treatment results.

RESUMO Objetivo: Com o objetivo de estudar a relação entre atividade aeróbia e atividade nervosa autonômica cardíaca por algoritmo de rede neural artificial e fusão biológica de imagens, tendo em vista os problemas existentes no modelo de rede neural artificial (RNA), é introduzida a tecnologia de processamento biológico de imagens com base em ANN. Métodos: um algoritmo de inteligência biológica de imagem Ann é proposto, um classificador adequado para triagem eletrocardiográfica (ECG) é projetado e um sistema de triagem de sinal de ECG é estabelecido com sucesso. Além disso, o conjunto de dados de sinais de ECG normais recuperados dos sujeitos durante o período experimental é construído e um classificador é usado para extrair os dados característicos de um sinal de ECG normal durante o período experimental. Resultados: As alterações na frequência cardíaca em repouso e outros indicadores de saúde física são analisadas pela combinação de indicadores fisiológicos de repouso, a saber, frequência cardíaca, peso corporal, índice de massa corporal e índice de gordura corporal. Os resultados mostram que o classificador autodesenhado pode processar com eficiência as imagens de ECG, e as atividades regulares de longo prazo podem melhorar as condições físicas da maioria das pessoas. O peso corporal e a taxa de gordura corporal da maioria dos indivíduos diminuem com a extensão do tempo do experimento, e a freqüência cardíaca em repouso diminui relativamente. Conclusões: Certos indicadores podem ser usados para prever a saúde física dinâmica de uma pessoa, o que indica que a pesquisa experimental de predição de índice nesta pesquisa tem um bom efeito, que não apenas estende a aplicação da rede neural artificial, mas também estabelece uma base para a pesquisa e implementação de dispositivos vestíveis de teste inteligente de ECG. Nível de evidência II; Estudos terapêuticos- investigação dos resultados do tratamento.

RESUMEN Objetivo: Para estudiar la relación entre la actividad aeróbica y la actividad del nervio autónomo cardíaco mediante el algoritmo de red neuronal artificial y la fusión de imágenes biológicas, ante los problemas existentes en el modelo de red neuronal artificial (ANN), se introduce la tecnología de procesamiento de imágenes biológicas basada en ANA. Métodos: Se propone un algoritmo de inteligencia de imagen biológica de Ann, se diseña un clasificador adecuado para el cribado electrocardiógrafo (ECG) y se establece con éxito un sistema de cribado de señales de ECG. Además, se construye el conjunto de datos de las señales de ECG recuperadas normales de los sujetos durante el período experimental, y se utiliza un clasificador para extraer los datos característicos de una señal de ECG normal durante el período experimental. Resultados: Los cambios en la frecuencia cardíaca en reposo y otros indicadores de salud física se analizan combinando indicadores fisiológicos en reposo, a saber, frecuencia cardíaca, peso corporal, índice de masa corporal y tasa de grasa corporal. Los resultados muestran que el clasificador de diseño propio puede procesar de manera eficiente las imágenes de ECG, y las actividades regulares a largo plazo pueden mejorar las condiciones físicas de la mayoría de las personas. El peso corporal y la tasa de grasa corporal de la mayoría de los sujetos disminuyen con la extensión del tiempo del experimento, y la frecuencia cardíaca en reposo disminuye relativamente. Conclusiones: Ciertos indicadores pueden usarse para predecir la salud física dinámica de una persona, lo que indica que la investigación experimental de predicción de índices en esta investigación tiene un buen efecto, lo que no solo extiende la aplicación de la red neuronal artificial sino que también sienta las bases para la investigación. e implementación de dispositivos portátiles de prueba inteligente de ECG. Nivel de evidencia II; Estudios terapéuticos- investigación de los resultados del tratamiento.

Research on athletes' psychological regulation ability based on back propagation (bp) neural network algorithm / Investigação sobre a habilidade de controle psicológico de atletas com base no algoritmo de rede neural backpropagation / Investigación sobre la capacidad de ajuste psicológico de los atletas basada en el algoritmo de red neuronal bp

ABSTRACT Athletes' psychological control ability directly affects competitions. Therefore, it is necessary to supervise the athletes' game psychology. Athletes' game state supervision model is constructed through the facial information extraction algorithm. The homography matrix and the calculation method are introduced. Then, two methods are introduced to solve the rotation matrix from the homography matrix. After the rotation matrix is solved, the method of obtaining the facial rotation angle from the rotation matrix is introduced. The two methods are compared in the simulation data, and the advantages and disadvantages of each algorithm are analyzed to determine the method used in this paper. The experimental results show that the model prediction accuracy reaches 70%, which can effectively supervise the psychological state of athletes. This research study is of great significance to improve the performance of athletes in competitions and improve the application of back propagation (BP) neural network algorithm.

RESUMO A capacidade de controle psicológico de atletas afeta diretamente as competições. Portanto, é muito necessário supervisionar a psicologia de jogo desses indivíduos. O modelo de supervisão do estado de jogo dos atletas é construído através do algoritmo de extração de informações faciais. A matriz de homografia e o método de cálculo são introduzido. Em seguida, são introduzidos dois métodos para resolver a matriz de rotação a partir da matriz de homografia. Após a resolução da matriz de rotação, introduz-se o método de obtenção do ângulo de rotação facial a partir dessa matriz. Os dois métodos são comparados nos dados da simulação, e as vantagens e desvantagens de cada algoritmo são analisadas para determinar o método utilizado neste estudo. Os resultados experimentais mostram que a precisão da previsão do modelo atinge 70%, sendo possível efetivamente supervisionar o estado psicológico dos atletas. O presente estudo é de grande importância para melhorar o desempenho dos atletas em competições e melhorar a aplicação do algoritmo de rede neural backpropagation (BP).

RESUMEN La capacidad de control psicológico de atletas afecta directamente las competencias. Por lo tanto, es muy necesario supervisar la psicología de juego de esos individuos. El modelo de supervisión del estado de juego de los atletas es construido por medio del algoritmo de extracción de informaciones faciales. La matriz de homografía y el método de cálculo son introducidos. Enseguida, son introducidos dos métodos para resolver la matriz de rotación a partir de la matriz de homografía. Después de la resolución de la matriz de rotación, se introduce el método de obtención del ángulo de rotación facial a partir de esa matriz. Los dos métodos son comparados en los datos de la simulación, y las ventajas y desventajas de cada algoritmo son analizadas para determinar el método utilizado en este estudio. Los resultados experimentales muestran que la precisión de la previsión del modelo alcanza 70%, siendo posible efectivamente supervisar el estado psicológico de los atletas. El presente estudio es de gran importancia para mejorar el desempeño de los atletas en competencias y mejorar la aplicación del algoritmo de red neuronal backpropagation (BP).

Demystifying artificial intelligence and deep learning in dentistry

Abstract Artificial intelligence (AI) is a general term used to describe the development of computer systems which can perform tasks that normally require human cognition. Machine learning (ML) is one subfield of AI, where computers learn rules from data, capturing its intrinsic statistical patterns and structures. Neural networks (NNs) have been increasingly employed for ML complex data. The application of multilayered NN is referred to as "deep learning", which has been recently investigated in dentistry. Convolutional neural networks (CNNs) are mainly used for processing large and complex imagery data, as they are able to extract image features like edges, corners, shapes, and macroscopic patterns using layers of filters. CNN algorithms allow to perform tasks like image classification, object detection and segmentation. The literature involving AI in dentistry has increased rapidly, so a methodological guidance for designing, conducting and reporting studies must be rigorously followed, including the improvement of datasets. The limited interaction between the dental field and the technical disciplines, however, remains a hurdle for applicable dental AI. Similarly, dental users must understand why and how AI applications work and decide to appraise their decisions critically. Generalizable and robust AI applications will eventually prove helpful for clinicians and patients alike.

The effects of ApoE epsilon4 alleles on cognitive function and resting-state functional MRI in patients with amnestic mild cognitive impairment: a prospective cohort study / 中华放射学杂志

Objective@#To explore the effects of ApoE epsilon4 (ApoE-ε4) alleles on cognitive function and resting-state functional MRI (rs-fMRI) in patients with amnestic mild cognitive impairment(aMCI) based on a prospective cohort study.@*Methods@#An average of 20 months of prospective observations were conducted on 16 ApoE-ε4-carriers and 24 non-carriers of aMCI. Neuropsychological assessments and rs-fMRI data were collected at both baseline and follow-up. All participants were assessed by a battery of neuropsychological tests and underwent rs-fMRI. Two core regions of the default mode network (DMN), the left posterior cingulate cortex (PCC) and the medial prefrontal cortex (mPFC), were selected as seeds to calculate the functional connectivity. Two-way repeated measures analysis of variance was used to assess the effects of ApoE genotype(ε4-carriers, nonε4-carriers), interval and the interaction between these two factors for functional connectivity extracted from changed region found by t-test.Conversion rates of dementia were compared between ApoE-ε4-carriers and nonε4-carriers at follow-up using Chi-square test. For the comparison of functional connectivity and clinical data between ApoE-ε4-carriers and nonsε4-carriers in baseline and follow-up, the normal distribution test was carried out first. If the normal distribution was fitted, the two-sample t test was used, otherwise, the Mann-Whitney rank sum test was used. Finally, the general linear model was used to assess the relationships between alterations in functional connectivity and in neuropsychological assessments as well as the interaction effect.@*Results@#(1)Significant decline in memory domains were found in ApoE-ε4-carriers as compared to non-carriers at both baseline and follow-up. The ApoE-ε4-carriers (14/16) presented a higher conversation rate than non-carriers(13/24, χ2=4.862, P=0.027) at follow-up. (2)Functional imaging analysis revealed that ApoE-ε4-carriers exhibited significantly higher functional connectivity between the left PCC and the left angular (ApoE-ε4-carriers: 0.23±0.11, non-carriers: -0.03±0.13, t=4.800, cluster size: 1 944 mm3, P=0.004), and between the left mPFC and the left angular (ApoE-ε4-carriers: 0.33±0.21, non-carriers: 0.08±0.18, t=5.040, cluster size:1 836 mm3, P=0.006) as compared to non-carriers at follow-up. We detected significant effect for the interaction interval by ApoE-ε4 on functional connectivity between the left angular and the left PCC (F=10.833, P=0.002)as well as the left mPFC (F=7.280, P=0.010). (3)The alteration of functional connectivity value between the left mPFC and the left angular in ApoE-ε4-carriers was positively correlated with the changes ofimmediate memory (r=0.692, P=0.018). The correlation was not statistically significant in ApoE-ε4-noncarriers (r=-0.198, P=0.417) and the integration effect was significant (F=8.632, P=0.006).@*Conclusions@#The ApoE-ε4 actually accelerates the deterioration of cognitive function in aMCI patients and carriers presented relatively reserved functional connectivity between the left angular and other core regions within DMN, which indicated the disruption of functional connectivity may be one of the underline mechanisms of ApoE-ε4 during AD process.

Research progresses of brain complex network in schizophrenia based on multimodal MRI / 中国医学影像技术

Schizophrenia (SZ) is a group of chronic mental disorders, which is often accompanied by perception, thinking, emotion, behavior and other impairments. MRI techniques can be used to investigate brain structural and functional alterations in SZ, so as to provide significant support for the recognition of biomarkers for mental disorders. Structural and functional brain networks in SZ constructed with multimodal MRI have been analyzed by the human brain connectome using graph theory in numerous studies, which highlighted the abnormality of brain complex networks, such as increased shortest path length, decreased clustering coefficient and global efficiency, as well as deficits of global hub, providing further support for the hypothesis of dysconnection in SZ. The recent advancements of structural networks, functional networks and multimodal networks were reviewed, and the characteristics of brain complex networks in SZ were explored, the existing problems of analysis methods and future direction were discussed in this paper.

Research progresses of convolutional neural network in musculoskeletal radiology / 中国医学影像技术

Artificial intelligence has been gradually applied in medical image diagnosis, showing good efficiency and diagnostic accuracy. As a recent innovation in artificial intelligence, convolutional neural network (CNN) displayed the ability to interpret medical images with accuracy at or near that of skilled clinicians for some applications, indicating overwhelming clinical application prospects. The research progresses of CNN in musculoskeletal radiology were reviewed in this article.