Date of Award

2016

Document Type

Thesis

Degree Name

Master of Science (MS) in Computer Science

Department

Computer Science

Abstract

"Automatically detecting early cancer using medical images is challenging, yet very crucial to help save millions of lives in the early stages of cancer. In this work, we improved a method that was originally developed by Yamaguchi et al. from the Saga University in Saga Japan. The original method would first decompose the endoscopic image into four color elements: red, green, blue and luminance (RGBL). Next each component is again decomposed to non-overlapping blocks of smaller images. Each smaller image undergoes two phases of DWT(s) and finally the Fractal Dimension (FD) is calculated per smaller image and abnormal regions are detectable. Our proposed method not only used GPU technology to speed up processing, this method also applied edge enhancement via Gaussian Fuzzy Edge Enhancement. After edge enhancement, multiple thresholds (or tuning variables) were identified and adjusted to reduce computational requirements, decrease false positives and increase the accuracy of detecting early cancer. Most lesions where a physician had manually indicated that could be an area of concern were detected quickly, less than four seconds, which is roughly 25x quicker than the existing work. The false positive rate was reduced but still needs improvement. In the future, a Support Vector Machine (SVM) would be an ideal solutions to reduce the false positive rate while also aiding in increasing detection and SVM technology has been implemented on the GPU. Once a technology, like a SVM, is implemented with better results, video processing will be the nearing the final step to 'Near Real Time Automatic Detection of Early Esophageal Cancer from an Endoscopic Image'"--Leaf iv.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Helms J 2016.zip (134420 kB)
Additional files

Share

COinS