Image Analysis And Treatment Of Duodenum - Gastric Reflux Ultrasonography

In clinical medicine, ultrasonography is widely used to detect Duodenogastric Reflux (DGR). Benefit from its good temporal resolution, low cost, noninvasiveness, no radiation and good repeatability, ultrasound plays an important role in the gastrointestinal examination. Up to now, clinicians have to check the ultrasonic video manually to identify DGR. The assessing process is tedious, time-consuming, and subject to the observer variability. In this dissertation, a novel and effective computer-aided method, including the region of interest (ROI) extraction, image velocimetry and automatic motion analysis, is proposed. It is meaningful to develop such a computer-aided analysis system to simplify the clinician’s work and perform the more quantitative and robust DGR evaluation.In the first part, the ROI is determined by an edge tracking strategy. By tracking gastroduodenal walls around pylorus, the pyloric region can be obtained, as well as the fluid region. Here an edge-based optical flow method is proposed. Combining optical flow with active contour, the proposed method improves its tracking accuracy by taking account of not only the spatio-temporal information between consecutive frames but also the connection constraint among edge points. The proposed method is compared with other four tracking methods using both synthetic and real ultrasonic image sequences. The experimental results show that the proposed method has the best agreement with the gold standard while having a stable tracking performance, which indicate that the proposed method outperforms the others.In the second part, an image velocimetry method based on weighted cross-correlation based variational optical flow is proposed to estimate the motion of pyloric fluid. Cross-correlation was combined with the variational optical flow framework as an illumination-robust local feature identifier. In consideration of accuracy near edges, the visual similarity weights according to the characteristics of ultrasonic images are constructed to improve the variational optical flow model, which makes the velocimetry assumption more reasonable. The proposed method is tested on synthetic and real ultrasonic images in comparison with other three optical flow methods. The results demonstrate that the proposed method is the most robust and accurate. It is a good tool for image velocimetry in ultrasonic images. In the third part, a system is proposed to realize the automatic motion analysis of B-mode ultrasonic videos and then the detection of DGR. After the ROI is determined, the motion field is estimated based on image velocimetry. Then an intelligent motion analysis is applied. A fluid-structure interaction analysis which combines both the motion and structural information is proposed to analyze the transploric motion of the fluid and finally determine the flow direction. The system is validated on ten B-mode ultrasonic images sequences, which consist of1000frames. The results show that automatic analysis is well accordant with the gold standard with the accuracy of0.959. The system can correctly detect the occurrence of DGR and get the duration of DGR. It is supposed to be a promising tool for the study and evaluation of DGR.