Segmentation of Kidneys Deformed by Nephroblastoma Using Case-Based Reasoning

Abstract

Image segmentation is a hot topic in image processing research. Most of the time, segmentation is not fully automated, and a user is required to guide the process in order to obtain correct results. Yet, even with programs, it is a time consuming process. In a medical context, segmentation can provide a lot of information to surgeons, but since this task is manual, it is rarely executed because of time,. Artificial Intelligence (AI) is a powerful approach to create viable solutions for a fully automated treatments. In this paper, we defined a Case-Based Reasoning (CBR) which can enhance region-growing segmentation of kidneys deformed by nephroblastoma. The main problem with region-growing methods is a user needs to place the seeds in the image manually. Automated methods exist but are not efficient every time and often give an over-segmentation. That is why we have designed an adaptation phase which can modify the cordinates of seeds recovered during the retrievial phase. We confronted our CBR approach with manual region growing and Convolutional Neural Network (CNN) to segment kidneys and tumors of CT-scans. Our CBR system succeded in performing the best segmentation for the kidney. keywords: Case-Based Reasoning Convolution Neural Network segmentation cancer tumour healthcare imaging artificial intelligence

A fusion method based on Deep Learning and Case-Based Reasoning which improves the resulting medical image segmentations

“…Each database has access to all the CT scan images and expert knowledge. The second layer is the segmentation layer, where CT scan images are segmented by artificial intelligence systems, specifically by a CBR system coupled with region growing technique for the pathological kidney (Marie et al (2018)) and by a Deep Learning segmentation, and more particularly by the FCN-8s for the nephroblastoma (Marie et al (2019)). At the end of the segmentation processes, the system gives two complementary segmentations in two different images and the fusion layer combines them with a neural network coupled with a CBR technique planned for the fusion of the two segmentations.…”

“…In the SAIAD platform, each structure is segmented separately by a technique of its own. Currently, the tumour is segmented by Deep Learning (DL) and more specifically by a Convolutional Neural Network (CNN) (Marie et al (2019)), and the pathological kidney is segmented by Cased-Based Reasoning (CBR) coupled with region growing technique (Marie et al (2018)), or by the same approach used for the segmentation of the tumour. However, once the segmentation step is done for each structure, the different segmentations must be aggregated together to obtain a final consensus segmentation.…”

Conflict management in the fusion of complementary segmentations of deformed kidneys and nephroblastoma

“…This 3D representation is currently done through manual segmentations, which is a time-consuming task. The French-Swiss border project SAIAD (Au- (Kato et al (2012)), and also the Fully Connected Conditional Random Fields (CRFs) Krähenbühl & Koltun (2011) coupled with CNNs to refine segmentations Chen et al (2018); Noh et al (2015); Kamnitsas et al (2017), genetic algorithm (Tosta et al (2017); Khan & Jaffar (2015)), and Case Based Reasoning (CBR) (Marie et al (2018); Kausar et al (2016); Frucci et al (2008)). But a single method is not efficient enough to achieve a correct segmentation of all structures on an image.…”

“…Each structure can be calculated using one optimal method of its own in order to obtain its best segmentation. Thus, in the SAIAD project, a method based on CBR and region growing is used in order to segment the kidney deformed by nephroblastoma on each of the patients' 2D images (Marie et al (2018)), and the OV 2 ASSION method, based on CNN, is used to segment nephroblastoma (Marie et al (2019)).…”