Neural Network Classifiers for Human Tissue Classification in NIR Biomedical Multispectral Imaging
| dc.contributor.author | Gurm, Sandeep | |
| dc.date.accessioned | 2018-06-11T15:23:13Z | |
| dc.date.available | 2018-06-11T15:23:13Z | |
| dc.date.issued | 2018-06-11 | |
| dc.date.submitted | 2018-06-05 | |
| dc.description.abstract | Near infrared imaging (NIR) is an imaging modality that has gained traction for solving biomedical problems in recent years. By leveraging the NIR spectrum, multiple spectra from the NIR range can be used to extract meaningful data from a variety of targets including human tissue; this technique is known as multispectral imaging (MSI) analysis. A generalized tissue classification method that identifies human tissue in an NIR multispectral imaging field is explored. NIR images are captured from four different wavelengths, and features are extracted from the individual images. The features are then manually labeled and used to train machine learning models to identify tissue/non-tissue areas within a multispectral image set. Although the application in this thesis is used to classify tissue/non-tissue, the techniques presented can be generalized to solve many other MSI classification problems in a variety of fields. In particular, two machine learning models are explored in this thesis; a multi-layer perceptron (MLP) and a convolutional neural network (CNN) approach. For each approach, feature selection and hyper-parameter tuning were used to design the machine learning architectures. After the design process, quantitative and qualitative tests were conducted to evaluate the merits of each algorithm design. Analysis found that the CNN approach yields excellent reliability and accuracy compared to the MLP. The accuracy, sensitivity, and specificity of the CNN is 95.2, 94.4, and 95.7% as calculated on a test set of MSI data. The MLP results on the same data set yield accuracy, sensitivity, and specificity values of 83.9, 85.4, and 83.1% respectively. It is also demonstrated that the CNN design maintains excellent accuracy even when challenged with varying tissue types and body compositions. The impact of this research will be most applicable to biomedical imaging modalities that utilize multispectral data. The techniques presented can be used to classify different types of tissues and their pathologies. Furthermore, the techniques can be generalized to other fields where multispectral data is used for inferencing, such as remote sensing applications. | en |
| dc.identifier.uri | http://hdl.handle.net/10012/13392 | |
| dc.language.iso | en | en |
| dc.pending | false | |
| dc.publisher | University of Waterloo | en |
| dc.subject | Neural networks | en |
| dc.subject | CNN | en |
| dc.subject | Multispectral imaging | en |
| dc.subject | Tissue classification | en |
| dc.subject | Biomedical imaging | en |
| dc.subject | NIR | en |
| dc.title | Neural Network Classifiers for Human Tissue Classification in NIR Biomedical Multispectral Imaging | en |
| dc.type | Master Thesis | en |
| uws-etd.degree | Master of Applied Science | en |
| uws-etd.degree.department | Systems Design Engineering | en |
| uws-etd.degree.discipline | System Design Engineering | en |
| uws-etd.degree.grantor | University of Waterloo | en |
| uws.contributor.advisor | Wong, Alexander | |
| uws.contributor.advisor | Badawy, Ossama | |
| uws.contributor.affiliation1 | Faculty of Engineering | en |
| uws.peerReviewStatus | Unreviewed | en |
| uws.published.city | Waterloo | en |
| uws.published.country | Canada | en |
| uws.published.province | Ontario | en |
| uws.scholarLevel | Graduate | en |
| uws.typeOfResource | Text | en |