Semiautomatic classification of benign and malignant vertebral fractures in magnetic resonance imaging

Abstract

The purpose of the study is the development of techniques for Computer Aided Diagnosis (CAD) for the identification and classification of vertebral compression fractures in magnetic resonance imaging (MRI). The proposal is to evaluate retrospectively the MR images of the lumbosacral spine performed in routine clinical practice at University Hospital in our institution. MRI images acquired T1-weighted sagittal will be used. Examinations of patients who underwent MRI of the lumbosacral spine will be included in the study. The feature extraction is obtained from a manual segmentation of the vertebral bodies. The benchmark used to test the performance of semi-automatic classification of vertebral fractures is a rank held by a senior radiologist specialist in diagnostic imaging of the spine and MR. The vertebral lumbar bodies area, coefficient of variation of the gray level histogram, skewness, kurtosis, moments and 14 Haralick attributes are the features that will be extracted. The feature vector is used as input to the classifier k-nearest and neural network.