Cascaded techniques for improving emphysema classification in computed tomography images

Abstract

The previous studies demonstrated the effectiveness of the multi-fractal based method for the classification of histo-pathological cases by calculating the local singularity coefficients of an image using different intensity measures. This paper proposed to improve the previous results by investigating the features derived from the combination of the alpha-histograms and the multifractal descriptors in the classification of Emphysema in computed tomography (CT) images. The performances of the classifiers are measured by using the classification accuracy (error matrix) and the area under the receiver operating characteristic curve (AUC). And further, the experimental results compared well with the local binary patterns (LBP) approach, a state-of-the-art measure for pulmonary Emphysema. The results also show that the proposed cascaded approach significantly improves the overall classification accuracy.

Identification of emphysema patterns in high resolution computed tomography images

“…Using the values of the intensity measures µ P (r) described in the studies, [6][7][8][20][21][22] we can explore how they scale with r and establish the power laws satisfied by them.…”

“…[ [6][7][8][9] Figure 3. Overview of emphysema pattern detection of the HRCT images using the Holder exponent computation There are several ways or methods of extracting the ROI in an image.…”

“…[ [6][7][8][9] High resolution computed tomography (HRCT) is a very effective scanning method that is more sensitive than chest radiography in measuring the type, extent and distribution of emphysema. Some patients with early emphysema, particularly those with early disease might still have the symptoms and this is the stage where HRCT is most useful.…”

“…[5][6][7][8][20][21][22] The major difference between the previous approaches and the proposed idea is that the previous work applied a global descriptor (multi-fractal spectrum) for identifying different areas in biomedical images, but this paper would use the local descriptor (Holder exponent) of the CT images for detecting the emphysema patterns. The computation of a multi-fractal descriptor depends on the calculated Holder exponent, which has been successfully used as a global descriptor for efficient classification in biomedical images, but since we are focusing on identifying the patterns within the images; the Holder exponent could be more appropriate for this task.…”

On Feature Selection Methods for Accurate Classification and Analysis of Emphysema CT Images

“…In this technique, the Holder exponent for the power law approximation of intensity measures in pixel neighbourhoods has been used for the computation of multi-fractal spectrum for the classification of images [5][6]. Detailed analysis of the emphysema classification using the multi-fractal techniques can be found in [7][8]. There are four different multi-fractal intensity measures: the summation, maximum, Iso and inverse minimum, which can be used for the computation of the Holder exponent but only the summation measure is considered for the experimental analysis in this paper.…”