Detection of Brain Tumor based on Features Fusion and Machine Learning

Abstract

Automated detection of brain tumor is a more challenging work due to the variability and complexity of shape, size, texture and location of lesions. The non-invasive MRI methods appear as a front line brain tumor detection tools (without ionization radiation). In this manuscript, an unsupervised clustering approach for tumor segmentation is proposed. Moreover, a fused feature vector is used which is a mixture of Gabor wavelet features (GWF), histograms of oriented gradient (HOG), local binary pattern (LBP) and segmentation based fractal texture analysis (SFTA) features. Random forest (RF) classifier is applied for classification between three sub tumoral regions such as complete, enhancing and non-enhancing tumor. To avoid over fitting problem, fivefold and 0.5 holdout cross-validation approaches are used. The promising detection efficiency depicts the dominance of proposed approach.

Appendix A

Notation	Description	Notation	Description
$\otimes$	Multiplication	${\text{p}}\left( {{\text{k}},{\text{l}}} \right)$	Matrix element
F	Input image	$\varphi ^{{\text{r}}}$	Gray level
L	Maximum image intensity	(+)	Fusion
P	Minimum image intensity	fs	Sinusoidal frequency
${\text{{M}}}$	Mean	$\theta$	Similar band orientation of a Gabor activity
${\text{{\rm A}}}$	Min	∅	Phase offset
${\text{{\rm B}}}$	Max	$\Sigma$	Standard deviation
$R$	Range	$\Gamma$	${\text{Spatial~aspect~proportion}}$
$\partial$	Interval	|G|	Gradient magnitude
${x_i}$	Foreground pixels	$\theta _{{\text{G}}}$	Gradient angle
N	Total number of pixels	$\in$	Small constant
${\text{f}}_{{\mu i}}$	Mean of the input image	V	non-normalized vector
${\text{f}}_{\gamma }$	The range of input image	${{\text{g}}_{\text{p}}}$	Gray level variance
${{\text{f}}_\partial }$	Interval of the input image	${\text{S}}$	Sum
${{\text{f}}_{{\vartheta _{\partial {\text{i}}}}}}$	Updated interval	${\text{P}},{\text{R}}$	Neighboring pixel
${\text{f}}_{\omega }$	Centroid of the input image	${\text{}}{{\text{g}}_{\text{c}}}$	Complementary contrast
${\text{f}}_{{\omega _{{\text{j}}} }}$	The difference of the centroid	$\nabla \left( {{\text{x}},{\text{y}}} \right)$	Tumor region boundaries
${\text{f}}_{{\vartheta _{{\omega {\text{j}}}} }}$	Updates centroid	N8[(x,y)]	8-connected pixel
${{\text{f}}_{{\text{binary}}}}$	Final segmented image	SE	Sensitivity
${{\text{f}}_{{\text{seg}}}}$	Color segmented image	SP	Specificity
$\varphi ^{{\text{f}}}$	Feature	ACC	Accuracy
$\varphi ^{{\mu {\text{f}}}}$	Mean feature	ROC	Receiver operating characteristic curve
$\varphi ^{{\sigma ^{2} }}$	Variance	DSC	Dice similarity coefficient
TPR	True positive rate	TP	True positive
FPR	False positive rate	FP	False positive
Q	Segmentation quality