宋镒凯,李时杰,梁月潇,班喜雷,吴 丹,夏明锋,陈丰荣.一种非酒精性脂肪性肝病的超声影像人工智能辅助诊断系统开发[J].,2023,(18):3401-3410 |
一种非酒精性脂肪性肝病的超声影像人工智能辅助诊断系统开发 |
Ultrasonography-based Artificial Intelligence System for the Classification and Quantification of Human Liver Fat Content |
投稿时间:2023-04-20 修订日期:2023-05-17 |
DOI:10.13241/j.cnki.pmb.2023.18.001 |
中文关键词: 非酒精性脂肪性肝病 超声检查 人工智能 肝脏脂肪含量 非侵入式肝脏肝诊断方法 |
英文关键词: Nonalcoholic fatty liver disease Ultrasonography Artificial intelligence Liver fat content Non-invasive method for fatty liver diagnosis |
基金项目:国家自然科学基金面上项目(32270724) |
|
摘要点击次数: 650 |
全文下载次数: 476 |
中文摘要: |
摘要 目的:结合人工智能方法设计针对肝脏超声影像的辅助诊断系统,辅助医生对大样本肝脏超声影像数据的标准化和高效化诊断,实现基于肝脏超声图像的非酒精性脂肪性肝病的精准诊断。方法:通过开发肝脏超声影像的识别与分类、脂肪肝分级分析和肝脏脂肪含量定量分析三个模块,建立一套非酒精性脂肪性肝病的超声影像人工智能辅助诊断系统,该系统能够自动区分输入到系统中不同采样视野的超声影像类型,并对肝脏超声图像进行数字化分析,给出待测超声图像是否呈现脂肪肝以及其肝脏脂肪含量的百分比值。结果:本研究中的超声图像识别分类模块可高通量区分出肝肾比图像和衰减率图像的两类超声影像,其分类的准确率达100%。脂肪肝分级分析模块在测试集数据的准确率达到84%,展现出可胜任辅助医生诊断的能力。基于人工肝脏脂肪含量定量方法开发的肝脏脂肪含量定量分析模块的准确率达到67.74%。结论:本研究已开发出一套基于肝脏超声影像的智能辅助诊断系统,可以辅助医生快速、简单、无创地筛选出潜在患有脂肪肝的患者,虽然现阶段实现肝脏脂肪定量分析仍有难度,但已展现出较大的临床应用潜力。 |
英文摘要: |
ABSTRACT Objective: To design an auxiliary diagnostic system for liver ultrasound images using artificial intelligence (AI) methods, to assist doctors in standardizing and efficiently diagnosing large samples of liver ultrasound images, and to achieve accurate diagnosis of non-alcoholic fatty liver disease (NAFLD) based on liver ultrasound images. Methods: By developing three modules: recognition and classification of liver ultrasound images, grading analysis of fatty liver, and quantitative analysis of liver fat content, a set of ultrasound image AI assisted diagnosis system for NAFLD is established. The system can automatically distinguish the types of ultrasound images with different sampling fields input into the system, perform digital analysis of liver ultrasound images, and provide whether the ultrasound image to be tested presents fatty liver and the percentage of liver fat content. Results: The ultrasound image recognition and classification module in this study could distinguish two types of ultrasound images with high throughput, namely hepatic/renal ratio images and hepatic attenuation rate images, with a classification accuracy of 100%. The accuracy of the fatty liver grading analysis module in the test set data reached 84%, demonstrating the ability to assist doctors in diagnosis. The accuracy of the liver fat content quantitative analysis module developed based on the artificial liver fat content quantitative method reached 67.74%. Conclusion: This study has developed an AI diagnostic system based on liver ultrasound imaging, which can assist doctors in quickly, simply, and non-invasive screening of potential patients with fatty liver. Although it is still difficult to achieve quantitative analysis of liver fat content so far, it has shown great clinical application potential. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |