Objective Animal wound model area measurement is a common biomedical research topic.Conventional manual measurement requires a lot of manpower to describe the contour,and there are still artificial errors in measurement.To propose an artificial intelligence algorithm measurement system based on VAT(volume aggregation with transformers).Methods A to-tal of 63 BALB/c mice were selected,which aged 9-10 weeks with body mass about 25 g.Among them,20 were simulated surgical cleansing wound models(treated with 5 mW,10 mW,15 mW light and natural healing),14 were simulated diabetic wound models(treated with 10 mW,15 mW light and natural healing)and 29 were simulated infectious wound models(treated with 5 mW,10 mW,15 mW light,antibiotics and natural healing).A total of 672 digital images were collected daily during wound healing.Based on image data,Image J software and artificial intelligence algorithm were used to delineate the wound area and calculate wound area data respectively.The comparative study between manual measurement and artificial intelli-gence algorithm was performed by analysis of image coincidence rate and wound area data.The manual measurements were completed by three people respectively.Results The effect of artificial intelligence algorithm and manual measurement had similar effects on image overlap,with highly positive correlation coefficient of area value r=0.968,P＜0.001,and intraclass correlation coefficient(ICC=0.824,P＜0.001),showing good consistency.The images were divided into 4 groups according to wound recovery process:open wound stage,scabbing stage,fiber healing stage and scar healing stage.Among them,the open wound stage and scab stages had r＞0.95,showing a highly positive correlation,fiber healing stage had r=0.490 and only scar healing stage had r=0.103.Conclusion It is demonstrated that artificial intelligence algorithm measurement system shows higher accuracy,stronger reliability and shorter analysis time than manual measurement,which has unique advantages in the area measurement of animal wound models.
NETL
NSTL
万方数据
