Semiautomatic classification of benign and malignant vertebral fractures in magnet...
| Grant number: | 15/08778-6 |
| Support type: | Scholarships abroad - Research Internship - Master's degree |
| Effective date (Start): | September 14, 2015 |
| Effective date (End): | December 20, 2015 |
| Field of knowledge: | Health Sciences - Medicine |
| Principal Investigator: | Marcello Henrique Nogueira-Barbosa |
| Grantee: | Lucas Frighetto Pereira |
| Supervisor abroad: | Rangaraj Mandayam Rangayyan |
| Home Institution: | Faculdade de Medicina de Ribeirão Preto (FMRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil |
| Local de pesquisa : | University of Calgary, Canada |
| Associated to the scholarship: | 14/12135-0 - Semiautomatic classification of benign and malignant vertebral fractures in magnetic resonance imaging, BP.MS |
Abstract The aim of this project is to apply specific medical image-processing techniques previously developed by the research team at the University of Calgary, to the computer-aided diagnosis system (CAD) of vertebral compression fractures (VCFs) in magnetic resonance imaging (MRI). This will provide a different analysis to complement the already implemented techniques in my post-graduation project "Semi-automatic classification of benign and malignant vertebral fractures in magnetic resonance imaging". As part of my Master´s project, we included 63 patients (38 women, 25 men, mean age 62.25 ± 14.13 years) with 103 vertebral bodies with VCFs and 106 without VCFs. Each one of these 209 vertebral bodies were manually segmented using the T1-weighted central sagittal slice lumbar spine MRI. F. To date we implemented image processing techniques to extract and analyze texture and contrast attributes. We evaluated the 14 attributes of Haralick and, as a contrast attributes, we used the calculation of coefficient of variation, skewness and kurtosis of gray level histogram. Tests using the classifier k-nearest neighbor (KNN) with k = 3, 10-folds cross-validation after selecting attributes reached an area under the ROC curve (AUROC) of 0.913 for the classification between benign and malignant VCFs. The techniques to be added on the current project are the calculation of compactness, convex deficiency, central invariant moments and Fourier descriptor of the vertebral bodies. These are shape analyses calculations and our hypothesis is that they may aid in the classification of the compression fractures of vertebral bodies. | |