| 张双双,陈笑一,石伟,王紫艺,韩童,杨鑫,倪东,黄兵旋,吴志霞,许娜.声像图质量对人工智能辅助诊断系统与人工测量发育性髋关节发育不良患儿生物学指标一致性的影响[J].中国医学影像技术,2024,40(7):1067~1071 |
| 声像图质量对人工智能辅助诊断系统与人工测量发育性髋关节发育不良患儿生物学指标一致性的影响 |
| Impact of ultrasonic image quality on the consistency of artificial intelligence assisted diagnosis system and manual measurements of biological indicators in children with developmental dysplasia of hip |
| 投稿时间：2023-12-09 修订日期：2024-02-03 |
| DOI：10.13929/j.issn.1003-3289.2024.07.022 |
| 中文关键词: 发育性髋关节发育不良 超声检查 人工智能 图像质量 |
| 英文关键词:developmental dysplasia of the hip ultrasonography artificial intelligence image quality |
| 基金项目:深圳市科技计划项目(JCYJ20220530160000001)、广东省高水平临床重点专科(深圳市配套建设经费)项目(SZGSP012)、深圳市“医疗卫生三名工程”项目资助(SZSM202011012)。 |
| 作者 | 单位 | E-mail | | 张双双 | 中国医科大学附属深圳市儿童医院超声科, 广东 深圳 518000 | | | 陈笑一 | 中国医科大学附属深圳市儿童医院超声科, 广东 深圳 518000 | | | 石伟 | 中国医科大学附属深圳市儿童医院骨科, 广东 深圳 518000 | | | 王紫艺 | 汕头大学医学院附属深圳市儿童医院超声科, 广东 深圳 518000 | | | 韩童 | 深圳大学医学部生物医学工程学院, 广东 深圳 518000 | | | 杨鑫 | 深圳大学医学部生物医学工程学院, 广东 深圳 518000 | | | 倪东 | 深圳大学医学部生物医学工程学院, 广东 深圳 518000 | | | 黄兵旋 | 中国医科大学附属深圳市儿童医院超声科, 广东 深圳 518000 | | | 吴志霞 | 中国医科大学附属深圳市儿童医院超声科, 广东 深圳 518000 | | | 许娜 | 中国医科大学附属深圳市儿童医院超声科, 广东 深圳 518000 | 46911069@qq.com |
|
| 摘要点击次数: 432 |
| 全文下载次数: 190 |
| 中文摘要: |
| 目的 观察声像图质量对人工智能(AI)辅助诊断系统与人工测量发育性髋关节发育不良(DDH)生物学指标一致性的影响。方法 回顾性分析75例DDH及345例非DDH患儿髋关节超声资料,对声像图进行主观评分。于每例选取1幅图像,以熵权法基于140幅(A组,取自25例DDH及115例非DDH)构建评估声像图质量模型,获得DDH相关解剖结构及影响因素的权重,据以将另外280幅 (B组,取自50例DDH及230例非DDH)图像质量由高至低分为A、B、C级;分析AI与人工测量B组DDH生物学指标的一致性。结果 DDH相关解剖结构及影响因素权重按照髂骨支下缘、髂骨、盂唇、骨缘转折点、股骨头、运动伪影依序下降。B组A级67幅(取自9例DDH及58例非DDH)、B级160幅 (取自26例DDH及134例非DDH)、C级53幅 (取自15例DDH及38例非DDH);除DDH患儿声像图中的β、股骨头覆盖率(FHC)及股骨头长径外,AI与人工测量DDH其他指标的一致性由高到低均为A级＞B级＞C级;二者针对DDH患儿声像图的测量结果的一致性高于非DDH。结论 髋关节声像图质量影响AI与人工测量DDH生物学指标的一致性;遇声像图质量欠佳时,应重点测量FHC及股骨头径线。 |
| 英文摘要: |
| Objective To observe the impact of ultrasonic image quality on the consistency of artificial intelligence (AI) assisted diagnosis system and manual measurements of biological indicators of developmental dysplasia of hip (DDH). Methods Hip ultrasonic data of 75 DDH and 345 non-DDH children were retrospectively analyzed, and the quality of ultrasonic images were subjectively scored. An evaluation model of ultrasonic image quality was constructed based on 140 ultrasonic images acquired from 140 cases (group A, containing 25 DDH and 115 non-DDH) using entropy weighting method, the weight of anatomic structures and impact factors related to DDH were obtained. The comprehensive image quality scores of other ultrasonic images acquired from 280 cases (group B, including 50 DDH and 230 non-DDH) were calculated, and the images in group B were classified into grade A, B and C in descending order. The consistency of AI and manual measurements of DDH biological indicators in group B was assessed. Results The weight of each anatomic structure and impact factors of DDH obtained with the model were as follows: The lower edge of iliac branch＞ilium＞glenoid labrum＞bony margin＞femoral head＞motion artifacts. In group B, grade A was observed in 67 (9 DDH and 58 non-DDH), grade B was found in 160 (26 DDH and 134 non-DDH), while grade C was noticed in 53 (15 DDH and 38 non DDH) images. Except for β, femoral head coverage (FHC) and femoral head length diameter, the consistencies between AI and manual measurements of other indicators of DDH were grade A＞B＞C. In group B, AI and manual measurements were more consistent in DDH than in non-DDH cases. Conclusion Ultrasonic image quality affected the consistency between AI and manual measurements of biological indicators of DDH. When image quality was not good enough, further attention should be paid to measurement of FHC and sizes of femoral head. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
