差分整合移动平均自回归模型与深度学习模型在吸脂操作数据预测分析中的应用比较

Comparison of autoregressive integrated moving average model and deep learning model in prediction and analysis of liposuction operation data

目的比较差分整合移动平均自回归模型(ARIMA)和深度学习模型在吸脂操作数据预测分析方面的应用价值。方法选取2019年1至9月中国医学科学院整形外科医院符合入选标准的行吸脂手术患者,使用基于光学追踪系统和力传感技术的吸脂操作记录系统,采集高年资整形外科医生吸脂手术初始250~400 s的操作数据,包括运动学和力学数据。经预处理后将采集数据分成一个吸脂往复循环为一组的数据。分别使用ARIMA模型和深度学习模型处理分析采集到的数据,建立吸脂操作预测模型。用Matlab 2017软件产生随机数随机抽取30对共计60组吸脂循环数据,计算每对数据的动态时间规整(DTW)值作为检验标准,然后分别计算基于ARIMA模型与深度学习模型的各30组预测数据与实际数据之间的DTW值,与检验标准对比,对2种模型的预测结果进行验证。应用Matlab 2017软件进行统计分析,2组比较用独立样本t检验,P<0.05为差异有统计学意义。结果共入组18例患者,均为女性,年龄23~49岁,平均36.6岁。吸脂部位分别为腹部、大腿、腰部。共获得16800组吸脂循环数据。模型检验标准DTW值为0.048±0.028。ARIMA模型预测数据与实际数据之间的DTW值为0.660±0.577,与检验标准比较差异有统计学意义(P<0.05)。深度学习模型得出的DTW值为0.052±0.030,与检验标准比较差异无统计学意义(P>0.05)。结论相比ARIMA模型,深度学习模型可以更准确地预测吸脂操作数据,能更好地适应不同情况的数据,并且具有更好的实时性。

Objective This study aims to compare the applicability value of autoregressive integrated moving average model(ARIMA)and deep learning model inprediction and analysis of liposuction operation data.Methods The patients who met inclusion criteria and underwent liposuction surgery in the Plastic Surgery Hospital of Chinese Academy of Medical Sciences from January 2019 to September 2019 were enrolled in this study.For each patient,250-400 s operation data including kinematics and mechanical data were collected by a senior plastic surgeon,using the liposuction operation recording system which consists of optical tracking and force sensing equipment.After pretreatment,the collected data were divided into one liposuction reciprocating cycle as one set of data.ARIMA model and deep learning model were used to analyze the collected data for establishing prediction models of liposuction operation.Using Matlab 2017,30 couples of liposuction data set were extracted by simple random sampling,and the dynamic time warping(DTW)value of each couple of data sets was calculated as test standard.Then,the DTW values between 30 sets of predicted data and actual data based on the ARIMA model and the deep learning model were calculated respectively and compared with the test standard to verify the prediction results of the two models.Matlab 2017 was used for statistical analysis.Independent sample t-test was used to compare the two groups,and P<0.05 indicated statistically significant difference.Results Eighteen patients were enrolled.All patients were females at 23-49 years old,with the mean age of 36.6 years old.Liposuction was performed in the abdomen,thighs,and waist.A total of 16800 sets of liposuction cycle data were obtained.The mean DTW value of test standard was 0.048±0.028.The mean DTW value between the ARIMA model predicted data and the actual data was 0.660±0.577,which was statistically significant compared with the test standard(P<0.05).The mean DTW value between the deep learning model predicted data and the actual data was 0.052±0.030,which was not significantly different compared to the test standard(P>0.05).Conclusions Compared with ARIMA model,deep learning model can predict liposuction operation data more accurately,and has better adaptability and real-time performance.