Premature Detection of Gastric lesions using Hybrid Deep Learning Techniques
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The gastrointestinal (GI) system of humans may be affected by a number of diseases, resulting in the deaths of around two million people worldwide. Endoscopy aids in the detection of these disorders, since the identification of these anomalies in endoscopic images of the GI tract is vital
newlinefor treatment and follow-up choices. The expansion of artificial intelligence as a consequence of the fast technological and scientific advancement has enhanced the detection of gastrointestinal diseases. The fundamental objective of this study is to classify gastric disorders from endoscopic
newlineimages using deep transfer learning methods and a proposed hybrid deep transfer learning method. However, doctors need sufficient time to examine such follow-ups, which inhibits manual diagnosis. This thesis presents a detailed evaluation of the relevant research on models for predicting gastric cancer. A literature study offers a detailed summary of the existing work on AI-enabled diagnostics for gastrointestinal malignancies at different stages. This thesis articulates the
newlinestatement of the research problem. Here, the researcher describes the issue they are attempting to tackle and the strategies they will use to do so. In addition, deep transfer learning algorithms and classification performance evaluation measures have been developed for evaluating the accuracy
newlineand precision of suggested gastric cancer predictions. This research evaluates many pre-trained models for loss, precision, root mean square error, precision, F1 score, and recall. Various transfer learning models and hybrid models are used and evaluated on the basis of a number of performance measures. This study utilised endoscopic images of a normal cecum, esophagitis, coloured resection margins, normal colon, normal Z-line, normal pylorus, ulcerative colitis, faeces, and polyps. Images are enhanced, and noise is removed, before being uploaded to deep
newlinetransfer learning algorithms. In addition, feature extraction techniques are used to get the region of interest by computing.