CAJ | 학술논문

进行脂肪肝超声图像的计算机辅助检测时，由医生手工选择感兴趣区域使得分类结果具有不确定性。为了避免人为因素对检测结果的影响，构造了一种利用图像分解技术确定ROI的算法，使得检测结果具有很强的稳健性。通过变分的方法把输入超声图像分解为纹理图和卡通图两部分，利用卡通图上的强度信息确定ROI，在纹理图上计算对应ROI的纹理特征。试验表明，相对于直接从超声图像上提取纹理信息，从分解后的纹理图像上提取的纹理信息有更好的分类效果。
진행지방간초성도상적계산궤보조검측시，유의생수공선택감흥취구역사득분류결과구유불학정성。위료피면인위인소대검측결과적영향，구조료일충이용도상분해기술학정ROI적산법，사득검측결과구유흔강적은건성。통과변분적방법파수입초성도상분해위문리도화잡통도량부분，이용잡통도상적강도신식학정ROI，재문리도상계산대응ROI적문리특정。시험표명，상대우직접종초성도상상제취문리신식，종분해후적문리도상상제취적문리신식유경호적분류효과。
When using computer aided detection system to detect fatty liver on ultrasonic liver image, ROI(Region Of Interest) selected by doctors leads to uncertain classification results. To avoid the influence which caused by artificial factor, an approach can get a robustness, ROI independent result for fatty liver detection is proposed by using image decomposition technique. Each liver image is decomposed into a cartoon part and a texture part by a variation method. The cartoon part is used to indicate the location of ROI while texture feature will be calculated in the according texture part. Experimental results show that the texture information extracted from the decomposition texture image has better classification accuracy than the one extracted from the original ultrasonic liver image.