Objective To compare the accuracy of18F-fluorodeoxyglucose(18F-FDG)PET in locating the epileptogenic focus in patients with temporal lobe epilepsy using the single-patient research method based on statistical parametric mapping(SPM)at dif-ferent confidence levels,and to compare it with the asymmetry index(AI)analysis method.Methods A retrospective analysis was conducted on the clinical and imaging data of 86 patients with drug-resistant temporal lobe epilepsy and 37 healthy controls.The pri-mary epileptogenic focus were located by 18F-FDG PET,and two-sample t-test and intracranial asymmetry analysis were performed on individual patients based on SPM.The accuracy of 18F-FDG PET in locating the epileptogenic focus was compared at different confidence levels P＜0.05,P＜0.05[familywise error rate(FWE)corrected],P＜0.01,P＜0.001,respectively.The diagnostic accuracy of the SPM method at the optimal confidence level P value was compared with the AI analysis method,and the data were analyzed using the x2 test.Results The diagnostic accuracy of the two-sample t-test method were 69.77％,79.07％,67.44％ and 63.95％ at confidence levels of P＜0.05,P＜0.05(FWE corrected),P＜0.01 and P＜0.001,respectively;the diagnostic accuracy of the intracranial asymmetry analysis method were 94.19％,81.39％,79.07％,and 75.58％,respectively.There was a statistically significant differ-ence in diagnostic accuracy between the intracranial asymmetry analysis method(P＜0.05)and the two-sample t-test method P＜0.05(FWE corrected)(x2=8.482,P＜0.05);there was also a statistically significant difference in AI analysis method between the two methods(x2=4.793,P＜0.05).Conclusion The intracranial asymmetry analysis method(P＜0.05)based on SPM has a higher accuracy in locating the primary epileptogenic focus in unilateral drug-resistant temporal lobe epilepsy than those in the two-sample t-test method P＜0.05(FWE corrected)and AI analysis method.

Objective To modify the YOLOv5 algorithm with similarity attention mechanism(SimAM)to enhance the recognition and segmentation accuracy of CT images for cerebral hemorrhage.Methods A basic framework was established with a YOLOv5 algorithm consisting of a backbone network(Backbone),a neck module(Neck)and a head module(Head),and then SimAM was introduced at the end of Backbone to form a YOLOv5-Sim-B algorithm and at the end of Neck to construct a YOLOv5-Sim-N algorithm.The YOLOv5-Sim-B and YOLOv5-Sim-N algorithms were trained and validated using the CT image dataset for cerebral hemorrhage publicly available on the Kaggle competition platform,and compared with the traditional YOLOv5 algorithm for recognizing and segmenting cerebral hemorrhagic lesions in CT images.Results In case the value of IoU-T was 0.6,the mean average precision(mAP)was 0.967 for YOLOv5-Sim-B algorithm,0.960 for the YOLOv5-Sim-N algorithm and 0.964 for the traditional YOLOv5 algorithm during the recognition and segmentation of cerebral hemorrhagic lesions in CT images.Conclusion The proposed algorithm gains advantages in detection accuracy and robustness,and can efficiently identify and segment cerebral hemorrhage foci in CT images.[Chinese Medical Equipment Journal,2025,46(5):1-8]

Objective To propose an improved YOLOv8 algorithm based on polarized self-attention(PSA)and deformable attention(DAT)so as to enhance the detection of pulmonary nodules in CT images.Methods A basic framework was established with a YOLOv8 model consisting of a backbone network(Backbone),a neck module(Neck)and a head module(Head).PSA was introduced into the end of the spatial pyramid pooling-fast(SPPF)of Backbone to construct a YOLOv8-PSA algorithm,and DAT was involved in the medium-scale feature layer P4 in Head to form a YOLOv8-DAT algorithm.The YOLOv8-PSA and YOLOv8-DAT algorithms were trained and validated using the CT image dataset of pulmonary nodules from public platforms,and compared with the original YOLOv8 algorithm for the detection of pulmonary nodule lesions in CT images.Results When used for pulmonary module detection of CT images,the YOLOv8-DAT algorithm had the mean average precision(mAP)in case of intersection over union threshold of 0.5(mAP50),mAP in case of intersection over union threshold of 0.5 to 0.95(mAP50-95)and precision ratio being 0.918,0.588 and 0.960 respectively,which gained advantages over the YOLOv8-PSA algorithm with mAP50,mAP50-95 and precision ratio being 0.914,0.583 and 0.945 respectively,and over the original YOLOv8 algorithm with mAP50,mAP50-95 and precision ratio being 0.911,0.564 and 0.952 respectively.Conclusion The YOLOv8-DAT algorithm detects pulmonary modules in CT images effectively,and facilitates early screening and diagnosis of pulmonary modules clinically.[Chinese Medical Equipment Journal,2025,46(8):1-10]

An overview of the development of machine learning in medical imaging was presented.The current status of machine learning applied to CT and MRI imaging diagnosis and prognosis prediction of hepatocellular carcinoma(HCC)was summarized.The challenges faced by machine learning in the application of CT and MRI imaging diagnosis and prognosis prediction of HCC were analyzed and its future development directions were envisioned.[Chinese Medical Equipment Journal,2025,46(7):92-98]

Objective To establish an electrical field(EF)stimulation model for Schwann cells(SCs),and to provide a basis for exploring the mechanisms of EF stimulation in promoting proliferation,migration and epithelial-to-mesenchymal transition of SCs.Methods A YC-3 bipolar programmable electrical stimulator and an electrotaxis chamber were used to construct an EF stimulation system to stimulate SCs.In the study,SCs were divided into control group(Ctrl)receiving no EF stimulation and EF group stimulated by continuous constant-voltage EF(100 mV/mm,3 h).The effects of EF stimulation on the proliferation and migration of SCs were analyzed using CCK-8 assay,and wound healing assay+Transwell assay,separately;and its effect on SCs adhesion was observed by analyzing the expressions of E-cadherin and N-cadherin using Western Blot.Results The CCK-8 assay results suggested that the absorbance at 450 nm was significantly higher in EF group than in Ctrl group(P<0.05).The results of wound healing assay+Transwell assay revealed that EF group had higher cell migration efficiency than Ctrl group(P<0.05).Western Blot results showed decreased E-cadherin expression and increased N-cadherin expression in EF group as compared with Ctrl group(P<0.05).Conclusion The improved EF stimulation system for SCs is operable.EF stimulation can promote the proliferation and migration of SCs.The decreased E-cadherin expression and increased N-cadherin expression may be related to the occurrence of epithelial-to-mesenchymal transition in SCs after EF stimulation.

The advantages of deep learning were introduced when appled to CT imaging,and the present situation of deep learning applied to cardiovascular CT imaging was reviewed in terms of image quality enhancement and de-noising,cardiac structure segmentation and quantitative measurement.The problems of deep learning and challenges encountered during the application to cardiovascular CT imaging were analyzed,and the future development directions included cross-institutional collaborative research,enhanced standardization of the data acquisition process and improved model interpretability.[Chinese Medical Equipment Journal,2025,46(10):97-105]

Objective To observe the value of machine learning(ML)models based on brain functional network features combining clinical indicators for predicting postoperative outcomes of patients with drug-resistant mesial temporal lobe epilepsy(DR-mTLE).Methods Totally 84 patients with unilateral DR-mTLE who underwent surgery were retrospectively enrolled and classified into seizure free(SF)group(n=55)and non-seizure free(NSF)group(n=29)according to one-year postoperative follow-up.Clinical data were analyzed to screen independent predictors of postoperative outcomes.Based on brain preoperative resting-state functional MRI,brain functional networks were constructed using graph theory analysis,and 587 features were extracted.Five-fold cross validation was used to divide the data into training set and test set,then the optimal brain functional network features related to postoperative outcomes of DR-mTLE patients were selected.Combining with clinically relevant independent predictors,ML models were constructed using classifiers including Gaussian process(GP),logistic regression(LR),support vector machine(SVM)and quadratic discriminant analysis(QDA),respectively,and the prediction efficacy,calibration and clinical value of each ML model were evaluated.Results Both course of disease and lesion location were clinically relevant independent predictors of postoperative outcome of DR-mTLE patients(OR=0.928,5.710,P=0.010,0.016).Four optimal brain function network features were selected,including betweenness centrality of the third zone of cerebellar vermis,degree centrality of right globus pallidus,nodal efficiency of temporal left inferior temporal gyrus and nodal clustering coefficient of left inferior parietal lobule.The average area under the curve(AUC)of GP,LR,SVM and QDA models in test set was 0.868,0.864,0.875 and 0.870,respectively.Calibration curves and decision curve analysis indicated that each ML model had good calibration and high clinical net benefit.Conclusion ML models based on brain functional network features combining with clinical indicators could be used to effectively predict postoperative outcomes in DR-mTLE patients.

Objective:To investigate the impact of immunosuppressive therapy on the severity of SARS-CoV-2 infection and cytokine levels in pediatric patients with kidney diseases.Methods:A retrospective analysis was conducted on the clinical data of 40 hospitalized pediatric patients who were diagnosed with SARS-CoV-2 infection at the 900th Hospital of PLA Joint Logistic Support Force from December 2022 to February 2023. Based on their immunosuppressive status prior to SARS-CoV-2 infection, these patients were categorized into immunosuppressive group and non-immunosuppressive group. Independent sample t-tests, Mann-Whitney U tests, and χ2 test were employed to compare the clinical baseline characteristics and laboratory data, the severity of SARS-CoV-2 infection, and the levels of cytokines between the 2 groups. Results:Among the 40 patients, 11 were in the immunosuppressive group (aged 13 (8, 14) years, 9 males and 2 females) and 29 in the non-immunosuppressive group (aged 2 (1, 4) years, 15 males and 14 females). In the immunosuppressive group, 2 were asymptomatic cases, 8 were mild cases, and 1 was moderate case, and there was no severe or critical cases. In the non-immunosuppressive group, 8 were mild cases, 5 were moderate, 15 were severe cases, 1 was critical case, and no asymptomatic cases. The underlying diseases in the immunosuppressive group included nephrotic syndrome (6 cases), IgA vasculitis nephritis (2 cases), lupus nephritis (1 case), post-renal transplantation (1 case), and renal failure (1 case), with a mean total immunosuppression score (TIS) of (3.6±1.4) points. In the non-immunosuppressive group, 2 patients had a history of epilepsy, and the remaining 27 cases had no underlying conditions, all with TIS scores of 0. Compared to the children in the non-immunosuppressive group, those in the immunosuppressive group were more likely to exhibit asymptomatic or mild infection, with lower risks of severe disease, cytokine storm, fever, and cough, but a higher risk of fatigue ( OR=1.22, 2.66, 0.48, 0.12, 0.12, 0.13, 1.22; 95% CI 0.93-1.62, 0.99-7.15, 0.33-0.70, 0.03-0.57, 0.03-0.57, 0.03-0.65, 0.93-1.62; all P<0.05). The levels of cytokine IL-6, interferon-α and interferon-γ in the immunosuppressive group were all lower than those in the non-immunosuppressive group ( Z=2.23, 2.51, 2.92, respectively; all P<0.05). Conclusion:Pediatric patients with kidney diseases receiving appropriate immunosuppressive therapy may mitigate the severity of SARS-CoV-2 infection by suppressing the expression of cytokines.

Objective To modify the YOLOv5 algorithm with similarity attention mechanism(SimAM)to enhance the recognition and segmentation accuracy of CT images for cerebral hemorrhage.Methods A basic framework was established with a YOLOv5 algorithm consisting of a backbone network(Backbone),a neck module(Neck)and a head module(Head),and then SimAM was introduced at the end of Backbone to form a YOLOv5-Sim-B algorithm and at the end of Neck to construct a YOLOv5-Sim-N algorithm.The YOLOv5-Sim-B and YOLOv5-Sim-N algorithms were trained and validated using the CT image dataset for cerebral hemorrhage publicly available on the Kaggle competition platform,and compared with the traditional YOLOv5 algorithm for recognizing and segmenting cerebral hemorrhagic lesions in CT images.Results In case the value of IoU-T was 0.6,the mean average precision(mAP)was 0.967 for YOLOv5-Sim-B algorithm,0.960 for the YOLOv5-Sim-N algorithm and 0.964 for the traditional YOLOv5 algorithm during the recognition and segmentation of cerebral hemorrhagic lesions in CT images.Conclusion The proposed algorithm gains advantages in detection accuracy and robustness,and can efficiently identify and segment cerebral hemorrhage foci in CT images.[Chinese Medical Equipment Journal,2025,46(5):1-8]

The advantages of deep learning were introduced when appled to CT imaging,and the present situation of deep learning applied to cardiovascular CT imaging was reviewed in terms of image quality enhancement and de-noising,cardiac structure segmentation and quantitative measurement.The problems of deep learning and challenges encountered during the application to cardiovascular CT imaging were analyzed,and the future development directions included cross-institutional collaborative research,enhanced standardization of the data acquisition process and improved model interpretability.[Chinese Medical Equipment Journal,2025,46(10):97-105]