Tumor classification using ultra-wideband (UWB) : late-time tumor response

• Main Author: Koh, Ling Ling.
• Other Authors: Soh Cheong Boon
• Format: Final Year Project
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Medical electronics
• Online Access: http://hdl.handle.net/10356/46319
• Institution: Nanyang Technological University

In this project, the feasibility of tumour classification using Ultra-Wideband (UWB) late–time tumour response is investigated. This report summarizes the findings and evaluates the accuracy of the technique by studying how the UWB late-time breast tumour backscatter responses are affected by the morphology of the tumour. The backscatter responses are then processed and classified using neural networks. Using the Geometrical Theory of Diffraction, the morphology information of the signals is embedded in the real part of a specific number of poles. The poles are derived using the singular value decomposition technique. These poles are then used as inputs for the neural networks classifiers. In this project, breast tumours have been successfully classified into either benign or malignant using only the late-time part of the UWB responses. Using a basic pattern recognition neural network with one hidden layer and tansig neurons, the student has achieved 82.33% accuracy for the homogenous environment and 82.67% accuracy for the heterogeneous environment. The results obtained are encouraging and show that it is possible to classify breast tumours based on only the late-time responses.