|
KMID : 1144120130030030190
|
|
Biomedical Engineering Letters
2013 Volume.3 No. 3 p.190 ~ p.197
|
|
|
Non-invasive estimation of intracranial pressure in traumatic brain injury (TBI) using fully-anisotropic Morlet wavelet transform and support vector regression
|
|
Aghazadeh Babak Seyed
Ansari Sardar
Pidaparti Ramana
Najarian Kayvan
|
|
|
Abstract
|
|
|
Purpose: This paper aims to estimate the intracranial pressure (ICP) in patients with traumatic brain injuries (TBI) noninvasively using directional features obtained from the texture of brain CT image and support vector regression (SVR) method.
Methods: A fully anisotropic Morlet wavelet transform is performed on brain CT images and optimal feature vectors have been extracted to classify the images into two groups of mild and severe TBI. Genetic algorithms with the fitness functions of support vector machines (SVM) classification accuracy rates have been used to find the optimal feature vector. Finally, SVR is implemented to estimate the ICP of patients with TBI. The results are compared to the ones obtained using Dual Tree complex wavelet transform based directional features.
Results: Features obtained from anisotropic continuous complex wavelet transform are shown to be effective in separating data from two classes of mild and severe TBI. The highest classification accuracy rate of 94.43 percent is achieved. Also, using SVR, the ICP estimation results demonstrate that the proposed algorithm yields excellent performance with a mean absolute error of 4.25 mmHg compared to Dual Tree complex wavelet transform features with the mean absolute error of 5.48 mmHg.
Conclusions: The severity of TBI is assessed non-invasively using brain CT images, and the directional textural features of brain tissue. The proposed algorithm using anisotropic Morlet wavelet features, GA-SVM based feature selection and SVR methods achieves an excellent performance in ICP estimation for TBI severity assessment.
|
|
|
KEYWORD
|
|
|
Intracranial pressure, Fully-anisotropic Morlet wavelet, Support vector regression
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|