Implementation of Cascaded CNN architecture for Fully automated Multiple modalities-based Brain tumour segmentation using selective overlapping patches

Shrivastava, Vibhash Anil Kumar (2022) Implementation of Cascaded CNN architecture for Fully automated Multiple modalities-based Brain tumour segmentation using selective overlapping patches. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (884kB) | Preview

Abstract

In this paper, the study aims to design & implement the fully automated High & low grade-based segmentation of glioblastomas brain tumors using Deep neural based CNN architecture. The proposed model utilized selective attention-based techniques which uses sizes of varying receptive field to identify the critical objects from the image in the successive layers of the CNN architecture. As a result, using the selective attention-based strategy in the CNN model provides us to extract the maximum number of appropriate features from the MRI images. The paper also addresses the two major challenges in the MRI images that is imbalance in class problems & identifying the spatial relationships amongst the patches in the images, for the foremost issue we suggested the uniform sampling method on Image patches(slice) & utilizing the class weighted based loss entropy function on the results of the segmentation for analysis. To address the second issue, we would be using the overlapping patches since it founds to be more effective in improving segmentation results against the adjacent patches as this would covers both the global and local features in the entire MRI images. The paper uses different modules of the CNN based architectures in order to maximize the feature extraction from the image with multiple skip connections with the modules. Further the results were evaluated on the basis of Dice score on different views and grades of tumor in the MRI image. The deep learning models were implemented using the BRATS 2018 datasets for the experimental analysis. The paper also utilizes the Unet architecture as an alternative methodology to segmentation process and compared with the proposed model in term of implementation time. Finally, the End-to-End automated deep learning-based CNN model shows the accurate and consistent segmentation results as compared to traditional methods which can be further utilized for the research proposed and clinical trials.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	CNN architecture; BRATS; class weighted; Overlapping patches; fully automated; Deep Learning model
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science R Medicine > RC Internal medicine > RC0254 Neoplasms. Tumors. Oncology (including Cancer) Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning Q Science > Life sciences > Medical sciences > Pathology > Tumors
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Tamara Malone
Date Deposited:	11 Mar 2023 13:02
Last Modified:	11 Mar 2023 13:02
URI:	https://norma.ncirl.ie/id/eprint/6307

Actions (login required)

View Item