A Bag-of-Features Approach to Classify Six Types of Pulmonary Textures on High-Resolution Computed Tomography

Abstract

SUMMARYComputer-aided diagnosis (CAD) systems on diffuse lung diseases (DLD) were required to facilitate radiologists to read highresolution computed tomography (HRCT) scans. An important task on developing such CAD systems was to make computers automatically recognize typical pulmonary textures of DLD on HRCT. In this work, we proposed a bag-of-features based method for the classification of six kinds of DLD patterns which were consolidation (CON), ground-glass opacity (GGO), honeycombing (HCM), emphysema (EMP), nodular (NOD) and normal tissue (NOR). In order to successfully apply the bag-of-features based method on this task, we focused to design suitable local features and the classifier. Considering that the pulmonary textures were featured by not only CT values but also shapes, we proposed a set of statistical measures based local features calculated from both CT values and eigenvalues of Hessian matrices. Additionally, we designed a support vector machine (SVM) classifier by optimizing parameters related to both kernels and the soft-margin penalty constant. We collected 117 HRCT scans from 117 subjects for experiments. Three experienced radiologists were asked to review the data and their agreed-regions where typical textures existed were used to generate 3009 3D volume-of-interest (VOIs) with the size of 32×32×32. These VOIs were separated into two sets. One set was used for training and tuning parameters, and the other set was used for evaluation. The overall recognition accuracy for the proposed method was 93.18%. The precisions/sensitivities for each texture were 96.67%/95.08% (CON), 92.55%/94.02% (GGO), 97.67%/99.21% (HCM), 94.74%/93.99% (EMP), 81.48%/86.03%(NOD) and 94.33%/90.74% (NOR). Additionally, experimental results showed that the proposed method performed better than four kinds of baseline methods, including two state-of-the-art methods on classification of DLD textures.

Blockwise Classification of Lung Patterns in Unsegmented CT Images

“…The overall results for CLBP SM C riu2 (8,2) can be seen in Table V. The classification results achieved by normal lung tissue, fibrosis and micronodules were similar to the literature, in which the micronodules category obtained the highest rate among some studies that used the same database (see [9]- [11]).…”

“…Figure 3 shows the magnified graphic for the ROC curves and allows to visualize the best feature extractor configuration. By looking at the overall results (Table III) and the ROC curves (Figure 3), one can see that the best results were achieved with CLBP SM C riu2 (8,2) , and the worst performance was with the CLBP SM C u2 (8,1) classifier. The remaining classifiers achieved very similar results.…”

“…The first was CLBP SM C u2 (P,R) that refers to the CLBP uniform pattern and the second was CLBP SM C riu2 (P,R) that is the rotation invariant uniform pattern. Both variations were applied using the values (8, 1) and (8,2) for the (P, R) parameters. Thus, four different CLBP combinations were tested.…”

“…After extracting the local features from the VOIs, an overcomplete dictionary was learned using Singular Value Decomposition (K-SVD) algorithm, and the descriptors were generated according to the dictionary and original feature vectors. Xu et al [8] presented a bag-of-features based method which combined both the original CT values and eigenvalues of Hessian matrix. A new classification method of lung tissues was presented by Song et al [9].…”

“…Definitions of the other metrics reported in the literature (sensitivity, accuracy and precision), as well as the specificity measure can be found in Section IV, since they were also used to evaluate this experiment. Anthimopoulos et al [5] Public -2503 ROIs Average F-score of 89% Dash et al [6] Private -100 ROIs Overall accuracy of 95% Zhao et al [7] Private -2360 VOIs Overall accuracy of 95.4% Xu et al [8] Private -3009 VOIs Overall accuracy of 93.18% Song et al [9] Public -23731 ROIs Overall precision of 80.7% Li et al [10] Public -16220 ROIs Overall sensitivity of 74.2% Depeursinge et al [11] Public -17848 ROIs Overall sensitivity of 76. 9% Malone et al [12] Private -852 ROIs Overall sensitivity of 89.5%…”