Clinical application evaluation of artificial intelligence-assisted semi-automatic segmentation software for liver hemangiomas
Objective:This study aimed to evaluate the performance of an artificial intelligence (AI)-assisted semi-automatic segmentation software in the segmentation of hepatic hemangiomas and compare it with traditional manual segmentation methods to determine if it can reduce intra-and inter-observer variability.Methods:A retrospective analysis was conducted on the clinical data of 52 patients with hepatic hemangiomas admitted from February 2022 to February 2023.A liver hemangioma analysis software developed by United Imaging Healthcare,based on deep learning technology,was used.Two physicians independently and blindly performed manual segmentation and AI-assisted semi-automatic segmentation of multi-slice spiral computed tomography (MSCT)images of the 52 hepatic hemangiomas.The longest diameter and volume were measured repeatedly using both manual and semi-automatic methods.Bland-Altman analysis was applied to assess intra-observer and inter-observer variability under both modes.The intraclass correlation coefficient (ICC)was used to evaluate the consistency of measurements between the two modes.Results:For the longest diameter measurement of liver hemangioma lesions,the mean deviations for intra-observer manual and semi-automatic measurements were-0.70 to 0.55 and-0.52 to 0.48,respectively,with 95％ limits of agreement.Inter-observer mean deviations were-0.15 to 1.24 for manual and-0.76 to 0.64 for semi-automatic measurements.For volume measurement,intra-observer mean deviations for manual and semi-automatic were-5.95 to 6.79 and-3.84 to 3.23,respectively,and inter-observer mean deviations were-8.04 to 5.56 for manual and-6.52 to 5.88 for semi-automatic.The ICC for intra-observer variability in the longest diameter measurement was 0.901 for manual and 0.976 for semi-automatic,and for inter-observer variability was 0.865 for manual and 0.892 for semi-automatic.For volume measurement,the ICC for intra-observer variability was 0.907 for manual and 0.982 for semi-automatic,and for inter-observer variability was 0.825 for manual and 0.913 for semi-automatic.The time required for semi-automatic lesion segmentation using AI-assisted software was significantly reduced compared to manual segmentation[(26.00±4.82)seconds vs.(7.23±2.89)seconds,t=-24.289,P<0.01].Conclusion:The AI-assisted semi-automatic segmentation software demonstrates accurate intra-and inter-observer consistency in the segmentation of liver hemangiomas.It significantly enhances segmentation efficiency,reduces the time required for clinical work,and has the potential to become a quantitative analysis tool for clinical follow-up and treatment efficacy assessment.
万方数据
维普
