NUST Institutional Repository

Segmentation and Classification of Gastroenterology Images using Saliency Mapping in CNNs

Show simple item record

dc.contributor.author Khan, Yusra
dc.date.accessioned 2024-08-30T07:09:48Z
dc.date.available 2024-08-30T07:09:48Z
dc.date.issued 2024-08-27
dc.identifier.issn 330106
dc.identifier.uri http://10.250.8.41:8080/xmlui/handle/123456789/46171
dc.description Supervisor: Dr. Arslan Shaukat en_US
dc.description.abstract This research work has been done towards the treatment of Gastrointestinal (GI) Cancer by experimenting as to how a computer aided design can help oncologists classify and segment GI Cancer using Gastroenterology (GE) Images. The concept of Attention Module by the name of Convolution Block Attention Module (CBAM) using Convolutional Neural Network (CNN) layers has been utilized to achieve saliency mapping in these images. The different layers of the CNN are tweaked and the hyper parameters are consistently checked and changed to achieve maximum accuracy. The CBAM Algorithm is divided into Channel Attention (CA) and Spatial Attention (SA) module. The Channel Attention has been implemented in the HSV Color Space for finding the best color space for segmentation purposes of the Chromo endoscopy medical images. Classification, followed by segmentation, is again implemented using CNN based model for normal and abnormal cancer images. For classification, 92% test accuracy has been achieved, while 70% Dice Coefficient has been obtained for segmentation. The results achieved are competitive as compared to the previous reported results on the GE images dataset. en_US
dc.language.iso en en_US
dc.publisher College of Electrical & Mechanical Engineering (CEME), NUST en_US
dc.subject Convolutional Block Attention Module (CBAM), CNN, GI Cancer Classification, GI Cancer Segmentation, HSV. en_US
dc.title Segmentation and Classification of Gastroenterology Images using Saliency Mapping in CNNs en_US
dc.type Thesis en_US


Files in this item

This item appears in the following Collection(s)

  • MS [332]

Show simple item record

Search DSpace


Advanced Search

Browse

My Account