Objective To understand the current situation of artificial intelligence production enterprise quality management system, so as to provide reference basis for the research and standardization of Artificial Intelligence Medical Device (AIMD) product quality management. Methods Based on YY/T 0287-2017 Medical Device Quality Management System for Regulatory Requirements, Medical Equipment Production and the Quality Control Standard for Independent Software Appendix and Xavier GMLP report, the relevant factors were screened and the questionnaire was designed by combining expert consultation and literature review. Then, a total of 32 representative AIMD enterprises were invited to fill in the questionnaire. Descriptive statistical analysis was performed on the data results using Excel 2016. Results Through in-depth analysis of the four themes in product planning and design, result output, product quality control and product change, it was found that it was necessary for enterprises participating in the survey to improve the quality management system of AIMD products to different degrees. Conclusions This study is the first time to systematically investigate the status quo of quality management of AIMD enterprises. The result will be useful for the establishment and continuous improvement of product quality management system. It will also provide a reference for the research of AIMD product quality management and the establishment of the standard.

目的 筛选出与人工智能医疗器械(Artificial Intelligence Medical Device,AIMD)企业质量管理体系相关的关键指标,从而为进一步构建生产研发过程中AIMD质量管理体系提供参考依据.方法 基于YY/T 0287-2017《医疗器械质量管理体系用于法规的要求》,通过两轮专家咨询法,筛选出与AIMD质量管理体系相关的指标,并对这些指标采用Likert 5点法进行权重打分.打分结果经过数据格式化处理后,应用IBM SPSS 21.0进行统计学分析,计算各指标的权重均值、标准差、变异系数(Coefficient of Variation,CV值)和满分率.结果 一共发放15份调研问卷,回收15份问卷,回收率为100％,专家的权威程度Cr值为0.91.共筛选得到5个一级指标、12个二级指标,36个三级指标,所有指标的平均分值介于3.40～4.93之间,标准差均小于1.00,CV值均小于25％.其中"设计和开发验证""设计和开发确认"和"设计和开发更改的控制"三个指标的得分均值分别为4.93、4.87和4.87,标准差分别为0.26、0.35和0.35,CV值分别为5.23％、7.23％和7.23％,满分率分别为93.33％、86.67％和86.67％.结论 通过两轮专家咨询法,基本确定了构建AIMD企业生产研发过程中的产品质量管理体系的关键指标,将为后续进一步构建产品质量管理体系提供参考依据.

Artificial intelligence is a blooming branch of medical device. Its development and quality control all rely on high quality clinical data. Since there is no established standard or guidance yet, it is important to study how to build and utilize a dataset appropriately and scientifically, especially for the decrease of clinical trial expense. With reference to the current status of premarket review and related guidance in developed countries, this paper analyzes the role and requirement of datasets in the quality control of AI medical device, providing useful information for regulation agencies and the development of public datasets for AI.

Electrocardiogram (ECG) measurement is an important part of wearable medical devices due to steady time of ECG signal acquisition and transmission through wireless system; therefore, reducing the amount of data transmission is very necessary. In this paper, the wearable medical devices ECG data processing based on compressive sensing theory is discussed. Based on the characteristics of ECG signal itself, the related operations with standard signal is proposed, so that the data can be compressed efficiently. The simulation results confirm the effectiveness of the algorithm.

Pulse response of radar system always suffers from high sidelobe level resulting in resolution degradation. Investigated here is a sidelobe suppression method based on apodization filtering technique for range responses of synthetic aperture radar (SAR) and noise SAR systems. The core of apodization filtering is finding an appropriate filtering vector in time domain. Compared with original apodization filtering, the proposed method could be realized stably because it could get correct filtering vector efficiently. This method contains three important steps: constructing coefficient matrix and desired response vector; performing ill-posed analysis; and solving equation to find filtering vector. In these steps, convolution kernel method is adopted to construct coefficient matrix; spectral condition is introduced as an indicating function for ill-posed analysis; and total variation method is used to resolve ill-posed equation for getting filtering vector. Elaborate theoretical derivation is presented to demonstrate the feasibility of this method. In order to test its effect, simulation experiments are implemented. Simulation results show that there is a great suppression of range sidelobes after processed by this method. With increasing filter length, the performance of filtered output is improved but time cost is increasing correspondingly. Furthermore, the proposed method is also effective with noise disturbance.

Grasping is a key operation of virtual hands. A grasping method for a virtual hand is presented to improve its precision in non-haptic environments. A motion control technique for the objects being or having been grasped is introduced by establishing a structural model of a virtual hand and the expression of the grasping process employing three states and their transitions. The penetrations between fingers and an object are solved by using the finger gestures adjustment strategy combined with an improved cyclic coordinate descent inverse kinematics algorithm and a forward kinematics algorithm. The grasping rules are established to ensure a true and accurate grasp. By integrating virtual devices into CATIA V5 using the secondary development platform CAA, a grasping examples for some simple objects in a virtual prototyping environment is developed, which demonstrates the correctness and real-time efficiency of the method.