Evaluation of Early Esophageal Adenocarcinoma Detection Using Deep Learning

Ghatwary, Noha and Ye, Xujiong and Zolgharni, Massoud (2018) Evaluation of Early Esophageal Adenocarcinoma Detection Using Deep Learning. In: CARS 2018 Computer Assisted Radiology and Surgery, 20-27, June, 2018, Berlin.

Documents
Evaluation of Early Esophageal Adenocarcinoma Detection Using Deep Learning

Request a copy
[img] PDF
Evaluation of EAC Detection Using Deep Learning.pdf - Whole Document
Restricted to Repository staff only

18MB
Item Type:Conference or Workshop contribution (Paper)
Item Status:Live Archive

Abstract

Esophageal Adenocarcinoma (EAC) is considered as the early stage of esophageal cancer developed mainly from the pre-malignant changes in esophagus lining named Barrett’s Esophagus (BE). Throughout the gastroin- testinal tract examination, premalignant and early cancer stages in the esoph- agus are usually overlooked as they are considered challenging to detect and requires a significant experience. Computer Aided Detection (CAD) systems, therefore, could be helpful in automatically detecting early cancerous lesions. With the recent advances in deep learning, the performance of object detec- tion methods has been increased to a great extent. In this paper, we aim to evaluate the performance of different state-of-the-art deep learning detection methods (RCNN, Fast, Rcnn, Faster RCNN, SSD) to automatically allocate BE abnormalities. To achieve that, a dataset of High-Definition white light en- doscopy images from 39 patients with corresponding manually annotated by five experienced clinicians has been evaluated. Experimental results show that Single Shor Multibox Detector (SSD) , outperforms other methods in terms of the evaluation measures

Keywords:Deep Learning, Esophageal Adenocarcinoma detection, Barrett’s Esophagus, HD-WLE
Subjects:G Mathematical and Computer Sciences > G400 Computer Science
Divisions:College of Science > School of Computer Science
ID Code:32583
Deposited On:05 Jul 2018 14:06

Repository Staff Only: item control page