You are here

Warning message

Attention! This event has already passed.

Development of a feature-deselective neuroevolution method and its relevance in medical CAD applications

Friday, 20 January, 2012 - 16:00
Campus: Brussels Humanities, Sciences & Engineering campus
Maxine Tan
phd defence

Computer-Aided Diagnosis (CAD) systems have been receiving increasing attention in
recent years. CAD has been defined as a diagnosis made by radiologists with the benefit of
information generated by computerized image analysis. The significant focus on developing
CAD systems for lung cancer has primarily been because it is by far the leading cause of cancer
death among both men and women in the United States. The primary reason for significant
optimism in CAD for lung cancer detection is the observation that those patients who are
diagnosed with early stage lung cancer, and who undergo curative resection have a much better
prognosis, with five-year survival rates rising to 40 to 70%. Many feature selection and
classification methods are proposed for lung cancer detection and diagnosis. A feature selection
method that automatically selects features at the same time as it evolves neural networks called
Feature Selective NeuroEvolution of Augmenting Topologies (FS-NEAT) was proposed by
Whiteson et al. In this work, we propose a novel feature selection method called Feature
Deselective NeuroEvolution of Augmenting Topologies (FD-NEAT), which begins with fullyconnected
inputs and automatically deselects irrelevant or redundant inputs. FD-NEAT, FSNEAT
and traditional NEAT are compared in some mathematical problems, and in a challenging
race car simulator domain (RARS). The results show that FD-NEAT outperforms FS-NEAT in
terms of network performance and feature selection, and evolves networks that offer the best
compromise between network size and performance. The performance of FD-NEAT is also
compared with that of two other established classifiers, namely support vector machines (SVMs)
and fixed-topology neural networks in a novel computer-aided lung nodule detection system for
computed tomography (CT) images. A set of 235 randomly-selected cases from the publiclyavailable
Lung Image Database Consortium (LIDC) database was used for training, and 125
independent cases for testing. The results show that the CAD system equipped with any of the
three classifiers performs well with respect to other methods described in literature.

PDF icon Tan_i.pdfPDF icon Tan_a.pdf