Liver And Tumor Segmentation Based On ResUNet

Biomedical image processing is a wide area in field of artificial intelligence&computer vision that provide complex and automated technology to get sophisticated performance and solution for problems statements.This is great significance for doctors to diagnosis diseases automatically using the advanced technologies and methods with accurate results.Segmentation of different medical images is the important part of medical images analysis,because medical image segmentation is used to diagnosis different diseases at its earlier stage for example segmentation of liver,and other human organs.Liver segmentation is an important problem because of increasing of liver cancer patient.Although there are many automated techniques have been developed for liver and tumor segmentation,however,segmentation of liver is still a difficult task due to the fuzzy&complex background of the liver position with other organs,complicated boundary and various appearance of liver.Therefore,developing a significance automatic liver and tumor segmentation from CT scans is very important for the analysis of liver cancer diagnoses.That’s why in this research topic we focused to developed trained a CNNs based model to segment the liver and liver’s tumor based on ResUNet.Firstly,we described about liver cancer disease,medical modalities to highlight the presence of tumor present in human body,and we described about different problems related to these modalities.Liver cancer is one of the most leading cause of cancer death in worldwide.Automatic liver segmentation from 3D biomedical images or CT scan is an essential challenging task for many clinical applications,such as surgical planning,hepatic diseases,postoperative analysis,and it can help doctors,radiologists to segment the liver tumor faster,and accurate analyses.Moreover,segmenting tumor from liver increases further dimensionality of difficulty because of overlap in intensity values and variability in position with other organs.Secondly,we highlight some of trained CNNs models,which are using for the liver segmentation and tumor segmentation like U-Net,ResNet-50 and VGGNet-19.But we checked that these CNNs trained models are still not providing enough information to segment liver and tumor because of its fuzzy boundaries and complicated structure.Thirdly,we used CNNs to overcome all the obstacles for segmentation of liver and tumor,we used ResUNet model on the 3D-IRCADb01 dataset by IRCAD which contains CT slices for patients along with masks for liver,tumors and other body organs.ResUNet is hybrid combination between the U-Net&ResNet,where it is uses Residual blocks rather than the traditional convolution blocks.In our techniques,we apply different strategies like 3-slice input;we used VGGNet-19,and RestNet-50 as base network and ResUNet method.We compare the combination of different strategies to check the final results of the liver and tumors separately.Our dataset contains 10 women and 10 men having hepatic tumors in 75%of cases ResUNet model consists of stacked layers of modified residual building blocks.We used the 2 cascaded CNNs approach one for segmenting the liver and extracting the ROI&the second one we use for extracted ROI from the first CNN and segment the tumors.We achieved dice coefficient of up to 95%and True value Accuracy of up to 99%and our comparison results show our method gives the best DICE score.Lastly,we high light research characteristics and innovations that explicit the proposed approaches provide better improvement,basis towards more effective loss functions,and Cross application for cross-datasets,which can be improved for segmentation of abdominal organs from any dataset’s modalities.Such as segmentation of heart and blood vessels from DICOM images,segmentation of liver from MRI,etc.by taking the advantages from ResUNet.