This study aimed to develop and validate a nomogram to forecast severe kidney disease (SKD) outcomes for hospitalized Henoch–Schönlein purpura (HSP) children. The predictive model was built based on a primary cohort that included 2,019 patients with HSP who were diagnosed between January 2009 and December 2013. Another cohort consisting of 461 patients between January 2014 and December 2016 was recruited for independent validation. Patients were followed up for 24 months in development/training and validation cohorts. The data were gathered at multiple time points after HSP (at 3, 6, 12, and 24 months) covering severe kidney disease as the severe outcome after HSP. The least absolute shrinkage and selection operator (LASSO) regression model was utilized to decrease data dimension and choose potentially relevant features, which included socioeconomic factors, clinical features, and treatments. Multivariate Cox proportional hazards analysis was employed to establish a novel nomogram. The performance of the nomogram was assessed on the aspects of its calibration, discrimination, and clinical usefulness. The nomogram comprised serious skin rash or digestive tract purpura, severe gastrointestinal (GI) manifestations, recurrent symptoms, and renal involvement as predictors of SKD, providing favorable calibration and discrimination in the training dataset with a C-index of 0.751 (95% CI, 0.734–0.769). Furthermore, it demonstrated receivable discrimination in the validation cohort, with a C-index of 0.714 (95% CI, 0.678–0.750). With the use of decision curve analysis, the nomogram was proven to be clinically useful. The nomogram independently predicted SKD in HSP and displayed favorable discrimination and calibration values. It could be convenient to promote the individualized prediction of SKD in patients with HSP.

Background Minimal change nephrotic syndrome (MCNS) is the most frequent cause of nephrotic syndrome in childhood. Previous studies have showed that respiratory syncytial virus (RSV) is the common trigger of MCNS. Methods Immunofluorescence of 3G10, 10E4 and HepSS1was performed in the kidneys, lungs and livers of RSV nephropathy rat model and the control. RSV fusion protein (RSVF) was detected by Western blot. The levels of HS domains in the kidneys and lungs of RSV nephropathy rat model were compared with that of RSVF. Results The expression of 10E4 and Hepss1 in kidney and lung of the normal rat was higher than 3G10. In the rat model of RSV nephropathy, RSVF of the kidneys and lungs showed stronger signal than the livers on day 8 and 14 after RSV infected. The expression of 3G10, 10E4 and Hepss1 in kidneys were obvious on day 4, then reduced from day 8 to 120. In the lungs the expression of 10E4 decreased continuously from day 8 to day 120, and Hepss1 reached the highest level on day 14 following by a reduction. While the level of RSVF was highest on day 8 when HS domains appeared weakly.Conclusions In the study, there were two main findings: (1) it was because of the Heterogeneity of HS, that is, rich N-sulfation in the kidney mediating RSV adhesion, which resulted in the progress of MCNS; (2) We firstly demonstrated that the damage directly of RSV and T-cell disorder in the pathogenesis of MCNS were equally important.

BACKGROUND Artificial intelligence (AI) has penetrated into almost every aspect of our lives and is rapidly changing our way of life. Recently, the new generation of AI taking machine learning and particularly deep convolutional neural network theories as the core technology, has stronger learning ability and independent learning evolution ability, combined with a large amount of learning data, breaks through the bottleneck limit of model accuracy, and makes the model efficient use. OBJECTIVE To identify the 100 most cited papers in artificial intelligence in medical imaging, we performed a comprehensive bibliometric analysis basing on the literature search on Web of Science Core Collection (WoSCC). METHODS The 100 top-cited articles published in “AI, Medical imaging” journals were identified using the Science Citation Index Database. The articles were further reviewed, and basic information was collected, including the number of citations, journals, authors, publication year, and field of study. RESULTS The highly cited articles in AI were cited between 72 and 1,554 times. The majority of them were published in three major journals: IEEE Transactions on Medical Imaging, Medical Image Analysis and Medical Physics. The publication year ranged from 2002 to 2019, with 66% published in a three-year period (2016 to 2018). Publications from the United States (56%) were the most heavily cited, followed by those from China (15%) and Netherlands (10%). Radboud University Nijmegen from Netherlands, Harvard Medical School in USA, and The Chinese University of Hong Kong in China produced the highest number of publications (n=6). Computer science (42%), clinical medicine (35%), and engineering (8%) were the most common fields of study. CONCLUSIONS Citation analysis in the field of artificial intelligence in medical imaging reveals interesting information about the topics and trends negotiated by researchers and elucidates which characteristics are required for a paper to attain a “classic” status. Clinical science articles published in highimpact specialized journals are most likely to be cited in the field of artificial intelligence in medical imaging.