基于B超图像的脂肪肝检测系统的研究

发布时间：2018-10-08 12:59

【摘要】：目前脂肪肝检查的手段主要是超声成像，但是由于超声成像存在一定的缺点和不足，需要借助一些计算机辅助手段才能达到较好的识别效果。而现有脂肪肝识别计算机辅助手段存在识别率不高和计算量大的缺点，而且不能满足现代化数字医院建设的要求。本文针对以上情况提出一种新的解决方法，对脂肪肝的识别和医院信息化建设起到了促进作用。在脂肪肝识别方面，本文针对两种不同类型的脂肪肝提出不同的解决方案。对于非均匀性脂肪肝采用图像增强的方法，结合直方图均衡和同态滤波的优点，对疑似区域性脂肪化的图片做增强处理，，以突出病变部位，提醒使用者注意；而对于弥漫型脂肪肝，分析对比了现有的脂肪肝计算机辅助识别手段以及它们的优缺点，提出一种基于B超图像的多级灰度差特征的识别方法。首先在感兴趣的肝脏图片上划定一个梯形区域，对划定区域中的血管影像噪声问题，采用改进的中值滤波算法进行去噪处理；然后将划定区域分成多个小的梯形区域，计算每个小梯形区域的平均灰度值，以得到肝脏图片多级灰度差特征；而后分析对比了正常肝和脂肪肝的多级灰度差特征曲线，确定多级灰度差特征的有效性；最后利用BP人工神经网络进行特征识别，达到图片区分的目的。在医院信息化建设方面，本文对支持DICOM3.0数字化接口和支持S端子模拟信号输出的B超检查设备，使用不同的图像采集方法，将脂肪肝检测系统和医院影像数据库连接，实现医院信息共享。特别针对医院病历报告的书写提出一种基于语义的报告单模板生成方法，提高了医生的工作效率。
[Abstract]:At present, the main method of fatty liver examination is ultrasonic imaging, but because of the shortcomings and shortcomings of ultrasound imaging, some computer-aided methods are needed to achieve a better recognition effect. However, the existing computer-aided methods for recognition of fatty liver have the disadvantages of low recognition rate and large amount of calculation, and can not meet the requirements of the construction of modern digital hospitals. This paper puts forward a new solution to the above situation, which can promote the recognition of fatty liver and the construction of hospital information. In recognition of fatty liver, this paper proposes different solutions for two different types of fatty liver. For non-uniform fatty liver, image enhancement method is adopted, combining the advantages of histogram equalization and homomorphism filtering, the image of suspected regional fatty liver is enhanced to highlight the lesion position and remind users to pay attention to it. For diffuse fatty liver, this paper analyzes and compares the existing computer-aided recognition methods and their advantages and disadvantages, and proposes a multi-level gray difference recognition method based on B-ultrasound image. Firstly, a trapezoid area is delineated on the liver image of interest, and the noise problem of vascular image in the delineated area is removed by using an improved median filtering algorithm. Then the delineated area is divided into a plurality of small trapezoid areas. The average grayscale value of each small trapezoid region is calculated to get the multilevel grayscale difference feature of liver picture, and then the multilevel gray difference characteristic curve of normal liver and fatty liver is analyzed and compared to determine the validity of multilevel grayscale difference feature. Finally, BP artificial neural network is used for feature recognition to achieve the purpose of image discrimination. In hospital information construction, this paper uses different image acquisition methods to connect the fatty liver detection system with the hospital image database, which supports the DICOM3.0 digital interface and the analog signal output of the S-terminal. To realize the sharing of hospital information. In particular, a semantic report template generation method is proposed for the writing of hospital medical records, which improves the efficiency of doctors.
【学位授予单位】：广西工学院
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R197.39;TP391.41

【参考文献】