Upper gastrointestinal anatomy detection with multi-task convolutional neural networks.

Esophagogastroduodenoscopy (EGD) has been widely applied for gastrointestinal (GI) examinations. However, there is a lack of mature technology to evaluate the quality of the EGD inspection process. In this Letter, the authors design a multi-task anatomy detection convolutional neural network (MT-AD-CNN) to evaluate the EGD inspection quality by combining the detection task of the upper digestive tract with ten anatomical structures and the classification task of informative video frames. The authors' model is able to eliminate non-informative frames of the gastroscopic videos and detect the anatomies in real time. Specifically, a sub-branch is added to the detection network to classify NBI images, informative and non-informative images. By doing so, the detected box will be only displayed on the informative frames, which can reduce the false-positive rate. They can determine the video frames on which each anatomical location is effectively examined, so that they can analyse the diagnosis quality. Their method reaches the performance of 93.74% mean average precision for the detection task and 98.77% accuracy for the classification task. Their model can reflect the detailed circumstance of the gastroscopy examination process, which shows application potential in improving the quality of examinations.

Study on the correlation between preoperative echocardiography indicators and postoperative prognosis in children with ventricular septal defect / 中华超声影像学杂志

Objective:To explore the correlation between preoperative echocardiography indicators and surgical prognosis of children with ventricular septal defect (VSD) and conduct verification based on significant indicators and indicator ratios.Methods:A total of 1 357 children with VSD who were admitted to the Children′s Hospital, Zhejiang University School of Medicine from June 2016 to June 2021 were selected. Various measurements including the size of the VSD, left ventricular ejection fraction (LVEF), left atrial (LA) diameter, the aortic (AO) flow rate, the tricuspid regurgitation velocity and pressure gradient were extracted from preoperative echocardiography reports. This paper explored the correlation between echocardiography reports indicators, indicator ratios and postoperative auxiliary ventilation time, respectively. The patients were divided into two groups according to whether there were complications, and the differences of echocardiography reports indicators between the two groups were compared. A linear regression model was established to predict the postoperative auxiliary ventilation time using these indicators, and the least absolute shrinkage and selection operator (LASSO) regression model was used for variable selection.Results:The VSD size and AO flow velocity were weakly correlated with the postoperative auxiliary ventilation time ( r=0.32, 0.25; all P<0.01). There was no significant correlation between VSD flow velocity and postoperative auxiliary ventilation time. The AO flow velocity/VSD flow velocity and LVEF/VSD flow velocity were strongly correlated with the postoperative auxiliary ventilation time ( r=0.67, 0.51; all P<0.01). In the significance test, there were no significant differences in tricuspid regurgitation flow velocity, tricuspid regurgitation pressure gradient, LA diameter, and LVEF between the complication group and the non-complication group(all P>0.01). However, the ratio of LVEF/tricuspid regurgitation velocity in the complication group was significantly lower than that in the non-complication group, and the ratio of tricuspid regurgitation pressure gradient/LA diameter was significantly higher than that in the non-complication group (all P<0.01). The postoperative auxiliary ventilation time of VSD patients was predicted on an independent test set, with an R2 of 0.51. Conclusions:Echocardiography report indicator ratios of AO flow velocity/VSD flow velocity and LVEF/VSD flow velocity have strong correlations with postoperative auxiliary ventilation time in children with VSD, and the ratios of LVEF/tricuspid regurgitation velocity and tricuspid regurgitation pressure gradient/LA diameter are significantly different between groups with and without postoperative complications. The ratios of indicators can significantly improve this correlation and difference, which can be used to predict the prognosis of VSD operation.

Establishment of normal reference system in pediatric echocardiography based on BigData / 中华超声影像学杂志

Objective To establish a pediatric echocardiographic normal reference system based on clinical BigData and overcome limitations such as insufficient sample size and diverse in methods of normalization . Methods Measurements were extracted from total 71 831 pediatric echocardiography reports in the past 5 years by using the Natural Language Processing ( NLP) technology . Among them ,a total of 12 732 reports were labeled as normal and were used to establish the normal reference system . A local regression ( LOESS ) approach was used to optimize both the reference value and variance across 5 grow th variables ( aortic diameter ,left atrium diameter ,left ventricle end‐diastolic endocardial diameter ,left main coronary artery diameter ,and right main coronary artery diameter) . T wo Z scores adjusted for age/sex and body surface area ( BSA ) were established respectively . In addition , 4 459 echocardiography reports with BSA information were used to evaluate these two Z scores . Results T wo Z scores generated from 4 459 reports showed pretty good normal distribution . T here were close strong correlations among two Z scores with Z scores generated based on the Pediatric Heart Network ( PHN ) . T he average correlation coefficient between BSA‐adjust Z scores and PHN Z scores was 0 .954 . T he average correlation coefficient between age/sex‐adjust Z scores and PHN Z scores was 0 .895 . T he results of this project were available as Z score calculator using the following link :http ://hdb .nbscn .org/zscore . Conclusions BigData provides a more efficient and better approach to establish normal reference systems in pediatric echocardiography .

Research situation and expectation of clinical healthcare big data / 医疗卫生装备

Objective To cxplore the research and application of big data mining technology in clinical healthcare environment.Methods The characteristics of clinical heahhcare data mining methods were expounded and the current research status of healthcare data mining technology was analyzed in clinical task according to the literature mining.Meanwhile,the application of healthcare data mining in clinical environment was introduced from the following four aspects:medicine research,personalized diagnosis,risk prediction and process mining.Results The necessity and importance of research and application of clinical healthcare data mining were illustrated.Moreover,the achicvements of healthcare data mining in clinical application,as well as existing problems and technical difficulties were summarized.Furthermore,the development tendencies of healthcare data mining were predicted.Conclusion The research and application of clinical healthcare data mining have improved the healthcare quality and promoted the medical advances,which is the developmcnt direction of clinical healthcare research.

Development method of healthcare information system integration based on business collaboration model / 生物医学工程学杂志

Integration of heterogeneous systems is the key to hospital information construction due to complexity of the healthcare environment. Currently, during the process of healthcare information system integration, people participating in integration project usually communicate by free-format document, which impairs the efficiency and adaptability of integration. A method utilizing business process model and notation (BPMN) to model integration requirement and automatically transforming it to executable integration configuration was proposed in this paper. Based on the method, a tool was developed to model integration requirement and transform it to integration configuration. In addition, an integration case in radiology scenario was used to verify the method.

Visual representation of digital clinical guideline / 生物医学工程学杂志

The digital clinical guidelines could greatly improve the safety and quality of clinical diagnosis and treatment. However, most of such guidelines were written in engineering language, which was difficult for clinicians to understand in practice. To tackle this problem, we adoped the flowchart as the visual representation method of digital clinical guidelines. The corresponding criterions expressed by the flowchart were easily understood by clinicians. Then we set the digital clinical guidelines written in Arden syntax as an example, the interconversion between flowcharts and digital clinical guidelines was realized. The result of using the visual representation method proposed in this paper shows that the clinical diagnosis logic becomes clear and intuitive. So this is an effective method for clinicians to understand and edit the digital clinical guidelines.

A research on healthcare integrating model of medical information system / 生物医学工程学杂志

System integration is inevitable since there are lots of heterogeneous medical information systems in the complicated medical environment. The current medical communication standards often focus on one aspect of the integration and do not provide a general scheme. Based on the analysis of the application of medical integration, the medical integration model HIM (Healthcare integrating model) is put forward, and the dataflow integration framework, function integration framework and interface integration framework in the HIM are designed subsequently. HIM provides a 3-D scheme for the integration of medical information systems, which not only contains the three aspects of integration application vertically, but covers the whole medical area horizontally.

Visualization of EIM simulation data on real cardiac model / 生物医学工程学杂志

Although 3D heart and torso model with realistic geometry are the basis of simulation computation in LFX Virtual Cardiac Model, the simulation results are mostly output in 2D format. Voxel mapping method is presented to solve this problem and extend the function of LFX Virtual Cardiac Model. Excitation Isochrone Map (EIM), one of the most important simulation results, was mapped from cardiac model with realistic geometry to real Visible Man cardiac model, then the EIM simulation data before and after mapping were visualized in the 4DView which is a real-time 3D medical image visualization platform. With this method, the output format of EIM simulation data of LFX Virtual Cardiac Model was extended from 2D to 4D (time is the 4th dimension) and from cardiac model with realistic geometry to real cardiac model. According to the EIM simulation data visualization results before and after mapping, the LFX virtual cardiac model shows its characteristics with more realistic and effective simulation.

A 3D real-time visualization system of medical image / 生物医学工程学杂志

Based on the integration of the advanced visualization toolkit (VTK) and the real-time rendering System (VolumePro), 4D View, a 3D real-time visualization system of medical image, which applies real-time 3D medical image (such as MR, CT image, etc.) visualization and interaction was developed. Firstly the VTK, VolumePro and their integration were introduced briefly; then the system design and function template of 4D View system was discussed in detail; finally some visualization results acquired through 4D View was illustrated. According to the results, 4D View system can effectively resolve the poor real-time characteristic of 3D medical image visualization and interaction, so it will have a wide application area of the clinical diagnosis, therapy and medical research, etc in future.