Improved Dempster and Shafer theory to fuse region and edge based level set for endocardial contour detection

Abstract

Data fusion is an important tool for improving the performance of a detection system when more than one classifier is available. The reasoning logic of DempsterShafer evidence theory for fusion is similar to that of humans. This paper discusses application ofa data fusion method which is based on improvements to the DempsterShafer theory, to echocardiographic images in order to increase the detection accuracy of the endocardial contours. In this paper, edge and region based level sets are implemented. The Improved Dempster-Shafer evidence fusion algorithm is applied to combine the detected contours resulting in promising results as shown by computational experiments.

Application of belief functions to medical image segmentation: A review

“…Similar to the BFT-based RGB medical image segmentation methods, the segmentation performance here is limited by feature extraction methods. Further research could build [127] MR images brain tumor segmentation EKNN Capelle et al [128] MR images brain tumor segmentation EKNN+Shafer's model + Appriou's model Barhoumi et al [129] optical imaging with color skin lesion malignancy tracking None Guan et al [130] MR images brain tissue segmentation Zhu's model Ketout et al [131] optical imaging with color endocardial contour detection Threshold Wen et al [132] MR images brain tissue segmentation Zhu's model+GD-based model George et al [133] optical imaging with color breast cancer segmentation Discounting Huang et al [96] PET-CT lymphoma segmentation ENN on recent advancements in deep feature representation and combine ECM with deep neural networks to learn mass feature representation. A good example of neural-network approach to evidential clustering can be found in [87].…”

“…Researchers in the medical domain have also recognized this limitation. In [131], Ketout et al proposed a modified mass computation method to address this limitation and applied their proposal to endocardial contour detection. First, the outputs of each active contour model (ACM) [136] were represented as mass functions.…”

Application of belief functions to medical image segmentation: A review

“…Thus we analyze the mass calculation methods in Table 4. There are some BFT-based segmentation method that use Bayesian instead of mass functions for information fusion (Barhoumi et al, 2007;Ketout et al, 2012;Huang et al, 2021a), or use discounting operation (George et al, 2020) or fuzzy mathematical morphology (Bloch, 2008) to calculate mass function (George et al, 2020), etc. For those methods, we summarized them into the 'Other' class.…”

“…Researchers in the medical domain has also awarded this limitation. Ketout et al (2012) proposed a modified mass computation method to break this limitation and apply their proposal for endocardial contour detection. First, the output of each active contour model (ACM) (Kass et al, 1988) was regraded as mass functions.…”

“…Similar to the autonomous driving task, the decision of segmentation and diagnosis in the medical domain should be cautious and reliable. Researchers in the medical have made some contributions to address the conflict between source information, such as the modified mass computation method (Ketout et al, 2012). At the same time, more contributions could be made to achieve the most reliable decision and to help the AI-based medical image diagnosis system applicable by taking reliability discounting into consideration.…”