Panayiotis Korfiatis

Texture-Based Identification and Characterization of Interstitial Pneumonia Patterns in Lung Multidetector CT

Identification and characterization of diffuse parenchyma lung disease (DPLD) patterns challenges computer-aided schemes in computed tomography (CT) lung analysis. In this study, an automated scheme for volumetric quantification of interstitial pneumonia (IP) patterns, a subset of DPLD, is presented, utilizing a multidetector CT (MDCT) dataset. Initially, lung-field segmentation is achieved by 3-D automated gray-level thresholding combined with an edge-highlighting wavelet preprocessing step, followed by a texture-based border refinement step. The vessel tree volume is identified and removed from lung field, resulting in lung parenchyma (LP) volume. Following, identification and characterization of IP patterns is formulated as a three-class pattern classification of LP into normal, ground glass, and reticular patterns, by means of k-nearest neighbor voxel classification, exploiting 3-D cooccurrence features. Performance of the proposed scheme in indentifying and characterizing ground glass and reticular patterns was evaluated by means of volume overlap (ground glass: 0.734 ± 0.057, reticular: 0.815 ± 0.037), true-positive fraction (ground glass: 0.638 ± 0.055, reticular: 0.942 ± 0.023) and false-positive fraction (ground glass: 0.361 ± 0.027, reticular: 0.147 ± 0.032) on five MDCT scans.

Texture classification‐based segmentation of lung affected by interstitial pneumonia in high‐resolution CT

Accurate and automated lung field (LF) segmentation in high-resolution computed tomography (HRCT) is highly challenged by the presence of pathologies affecting lung borders, also affecting the performance of computer-aided diagnosis (CAD) schemes. In this work, a two-dimensional LF segmentation algorithm adapted to interstitial pneumonia (IP) patterns is presented. The algorithm employs k-means clustering followed by a filling operation to obtain an initial LF order estimate. The final LF border is obtained by an iterative support vector machine neighborhood labeling of border pixels based on gray level and wavelet coefficient statistics features. A second feature set based on gray level averaging and gradient features was also investigated to evaluate its effect on segmentation performance of the proposed method. The proposed method is evaluated on a dataset of 22 HRCT cases spanning a range of IP patterns such as ground glass, reticular, and honeycombing. The accuracy of the method is assessed using area overlap and shape differentiation metrics (d(mean), d(rms), and d(max)), by comparing automatically derived lung borders to manually traced ones, and further compared to a gray level thresholding-based (GLT-based) method. Accuracy of the methods evaluated is also compared to interobserver variability. The proposed method incorporating gray level and wavelet coefficient statistics demonstrated the highest segmentation accuracy, averaged over left and right LFs (overlap=0.954, d(mean)=1.080 mm, d(rms)=1.407 mm, and d(max)=4.944 mm), which is statistically significant (two-tailed students t test for paired data, p<0.0083) with respect to all metrics considered as compared to the proposed method incorporating gray level averaging and gradient features (overlap=0.918, d(mean)=2.354 mm, d(rms)=3.711 mm, and d(max)=14.412 mm) and the GLT-based method (overlap=0.897, d(mean)=3.618 mm, d(rms)=5.007 mm, and d(max)=16.893 mm). The performance of the three segmentation methods, although decreased as IP pattern severity level (mild, moderate, and severe) was increased, did not demonstrate statistically significant difference (two-tailed students t test for unpaired data, p>0.0167 for all metrics considered). Finally, the accuracy of the proposed method, based on gray level and wavelet coefficient statistics ranges within interobserver variability. The proposed segmentation method could be used as an initial stage of a CAD scheme for IP patterns.

Vessel Tree Segmentation in Presence of Interstitial Lung Disease in MDCT

The automated segmentation of vessel tree structures is a crucial preprocessing stage in computer aided diagnosis (CAD) schemes of interstitial lung disease (ILD) patterns in multidetector computed tomography (MDCT). The accuracy of such preprocessing stages is expected to influence the accuracy of lung CAD schemes. Although algorithms aimed at improving the accuracy of lung fields segmentation in presence of ILD have been reported, the corresponding vessel tree segmentation stage is under-researched. Furthermore, previously reported vessel tree segmentation methods have only dealt with normal lung parenchyma. In this paper, an automated vessel tree segmentation scheme is proposed, adapted to the presence of pathologies affecting lung parenchyma. The first stage of the method accounts for a recently proposed method utilizing a 3-D multiscale vessel enhancement filter based on eigenvalue analysis of the Hessian matrix and on unsupervised segmentation. The second stage of the method is a texture-based voxel classification refinement to correct possible over-segmentation. The performance of the proposed scheme, and of the previously reported technique, in vessel tree segmentation was evaluated by means of area overlap (previously reported: 0.715 ± 0.082, proposed: 0.931 ± 0.027), true positive fraction (previously reported: 0.968 ± 0.019, proposed: 0.935 ± 0.036) and false positive fraction (previously reported: 0.400 ± 0.181, proposed: 0.074 ± 0.031) on a dataset of 210 axial slices originating from seven ILD affected patient scans (used for performance evaluation out of 15). The pro posed method demonstrated a statistically significantly (p <; 0.05) higher performance as compared to the previously reported vessel tree segmentation technique. The impact of suboptimal vessel tree segmentation in a reticular pattern quantification system is also demonstrated.

Automated 3D segmentation of lung fields in thin slice CT exploiting wavelet preprocessing

Lung segmentation is a necessary first step to computer analysis in lung CT. It is crucial to develop automated segmentation algorithms capable of dealing with the amount of data produced in thin slice multidetector CT and also to produce accurate border delineation in cases of high density pathologies affecting the lung border. In this study an automated method for lung segmentation of thin slice CT data is proposed. The method exploits the advantage of a wavelet preprocessing step in combination with the minimum error thresholding technique applied on volume histogram. Performance averaged over left and right lung volumes is in terms of: lung volume overlap 0.983 ±0.008, mean distance 0.770 ± 0.251 mm, rms distance 0.520 ± 0.008 mm and maximum distance differentiation 3.327 ± 1.637 mm. Results demonstrate an accurate method that could be used as a first step in computer lung analysis in CT.

Size-adapted microcalcification segmentation in mammography utilizing scale-space signatures

The purpose of this study is size-adapted segmentation of individual microcalcifications in mammography, based on microcalcification scale-space signature estimation, enabling robust scale selection for initialization of multiscale active contours. Segmentation accuracy was evaluated by the area overlap measure, by comparing the proposed method and two recently proposed ones to expert manual delineations. The method achieved area overlap of 0.61+/-0.15 outperforming statistically (p<0.001) the other two methods (0.53+/-0.18, 0.42+/-0.16). Only the proposed method performed equally for both small (< 460 microm) and large (>/= 460 microm) microcalcifications. Results indicate an accurate method, which could be utilized in computer-aided diagnosis schemes of microcalcification clusters.

Evaluating image denoising methods in myocardial perfusion single photon emission computed tomography (SPECT) imaging

The statistical nature of single photon emission computed tomography (SPECT) imaging, due to the Poisson noise effect, results in the degradation of image quality, especially in the case of lesions of low signal-to-noise ratio (SNR). A variety of well-established single-scale denoising methods applied on projection raw images have been incorporated in SPECT imaging applications, while multi-scale denoising methods with promising performance have been proposed. In this paper, a comparative evaluation study is performed between a multi-scale platelet denoising method and the well-established Butterworth filter applied as a pre- and post-processing step on images reconstructed without and/or with attenuation correction. Quantitative evaluation was carried out employing (i) a cardiac phantom containing two different size cold defects, utilized in two experiments conducted to simulate conditions without and with photon attenuation from myocardial surrounding tissue and (ii) a pilot-verified clinical dataset of 15 patients with ischemic defects. Image noise, defect contrast, SNR and defect contrast-to-noise ratio (CNR) metrics were computed for both phantom and patient defects. In addition, an observer preference study was carried out for the clinical dataset, based on rankings from two nuclear medicine clinicians. Without photon attenuation conditions, denoising by platelet and Butterworth post-processing methods outperformed Butterworth pre-processing for large size defects, while for small size defects, as well as with photon attenuation conditions, all methods have demonstrated similar denoising performance. Under both attenuation conditions, the platelet method showed improved performance with respect to defect contrast, SNR and defect CNR in the case of images reconstructed without attenuation correction, however not statistically significant (p > 0.05). Quantitative as well as preference results obtained from clinical data showed similar performance of the denoising methods studied. In conclusion, the multi-scale platelet denoising method applied on raw projection images provides more efficient noise reduction while preserving image quality in a myocardial phantom SPECT imaging as compared to the Butterworth filter applied either on projection or reconstructed images. However, this trend in favour of the platelet denoising method was not observed on clinical data reconstructed either without or with attenuation correction.

Towards quantification of interstitial pneumonia patterns in lung multidetector CT

Quantification of Diffuse Parenchyma Lung Disease (DPLD) patterns challenges computer aided diagnosis schemes in Computed Tomography (CT) lung analysis. In this study, an automated scheme for volumetric quantification of Interstitial Pneumonia (IP) patterns, a subset of DPLDs, is presented, utilizing a MultiDetector CT (MDCT) data set. Initially, Lung Field (LF) segmentation is achieved by 3D automated gray level thresholding combined to wavelet highlighting, followed by a texture based border refinement step. The vessel tree volume is identified and removed from LF, resulting in Lung Parenchyma (LP) volume. Following, the abnormal LP is differentiated from normal LP utilizing a 2 class k-means clustering. Quantification of IP patterns is formulated as a three-class pattern recognition problem to classify abnormal LP into ground glass, reticular and honeycomb patterns, by means of SVM voxel classification, exploiting 3D co-occurrence features. Performance of the proposed scheme in segmenting LF, as well as in quantifying normal LP, ground glass, reticular and honeycomb patterns was evaluated by means of volume overlap on 5 MDCT scans. Volume overlap for left LF and right LF was 0.95 plusmn 0.03 and 0.96 plusmn 0.02 respectively, while for normal LP, ground glass, reticular and honeycombing patterns was 0.89 plusmn 0.02, 0.70 plusmn 0.04, 0.72 plusmn 0.05 and 0.71 plusmn 0.03, respectively.

Automated vessel tree segmentation: Challenges in computer aided quantification of diffuse parenchyma lung diseases

Identification and characterization of diffuse parenchyma lung disease patterns challenges Computer Aided Diagnosis (CAD) schemes in Computed Tomography (CT). Accuracy of these preprocessing stages is expected to influence the accuracy of lung CAD schemes. Although algorithms aimed at improving the accuracy of segmentation of lung fields in presence of DPLDs have been reported, the corresponding vessel tree segmentation stage is under-researched. In this paper, an automated vessel tree segmentation scheme is proposed, utilizing a 3D multi-scale vessel segmentation filter based on eignen value analysis of the Hessian matrix and unsupervised segmentation, followed by texture classification refinement to correct possible over-segmentation. Performance of the proposed scheme in vessel tree segmentation was evaluated by means of volume overlap (no refinement: 0.794, refinement: 0.925), true positive fraction (no refinements: 0.938, refinement: 0.902) and false positive fraction (no refinement: 0.241, refinement: 0.077) to pixel exact ground truth of 3 MDCT scans.

Selecting registration schemes in case of interstitial lung disease follow-up in CT

PURPOSE Primary goal of this study is to select optimal registration schemes in the framework of interstitial lung disease (ILD) follow-up analysis in CT. METHODS A set of 128 multiresolution schemes composed of multiresolution nonrigid and combinations of rigid and nonrigid registration schemes are evaluated, utilizing ten artificially warped ILD follow-up volumes, originating from ten clinical volumetric CT scans of ILD affected patients, to select candidate optimal schemes. Specifically, all combinations of four transformation models (three rigid: rigid, similarity, affine and one nonrigid: third order B-spline), four cost functions (sum-of-square distances, normalized correlation coefficient, mutual information, and normalized mutual information), four gradient descent optimizers (standard, regular step, adaptive stochastic, and finite difference), and two types of pyramids (recursive and Gaussian-smoothing) were considered. The selection process involves two stages. The first stage involves identification of schemes with deformation field singularities, according to the determinant of the Jacobian matrix. In the second stage, evaluation methodology is based on distance between corresponding landmark points in both normal lung parenchyma (NLP) and ILD affected regions. Statistical analysis was performed in order to select near optimal registration schemes per evaluation metric. Performance of the candidate registration schemes was verified on a case sample of ten clinical follow-up CT scans to obtain the selected registration schemes. RESULTS By considering near optimal schemes common to all ranking lists, 16 out of 128 registration schemes were initially selected. These schemes obtained submillimeter registration accuracies in terms of average distance errors 0.18 ± 0.01 mm for NLP and 0.20 ± 0.01 mm for ILD, in case of artificially generated follow-up data. Registration accuracy in terms of average distance error in clinical follow-up data was in the range of 1.985-2.156 mm and 1.966-2.234 mm, for NLP and ILD affected regions, respectively, excluding schemes with statistically significant lower performance (Wilcoxon signed-ranks test, p < 0.05), resulting in 13 finally selected registration schemes. CONCLUSIONS Selected registration schemes in case of ILD CT follow-up analysis indicate the significance of adaptive stochastic gradient descent optimizer, as well as the importance of combined rigid and nonrigid schemes providing high accuracy and time efficiency. The selected optimal deformable registration schemes are equivalent in terms of their accuracy and thus compatible in terms of their clinical outcome.

Computer Aided Diagnosis of Diffuse Lung Disease in Multi-detector CT - Selecting 3D Texture Features

Computed Tomography (CT) is the modality of choice for the diagnosis of Diffuse Lung Disease (DLD) affecting lung parenchyma. The need for Computer Aided Diagnosis (CAD) schemes aimed at DLD patterns quantification in lung CT, originates from large inter- and intra observer variability characterizing DLD interpretation. The majority of the proposed CAD schemes aimed at DLD characterization exploits textural features combined with supervised classification algorithms. However, the exploitation of these features is suboptimal, since no feature reduction or evaluation is performed prior to the classification task. The aim of the current paper is to investigate 3D texture features sets (histogram signatures, co-occurrence and run length matrices’ statistics) regarding their capability in DLD patterns’ characterization (normal, ground glass, reticular and honeycombing). Earth Mover’s Distance (EMD), k-Nearest Neighbor (k-NN) and Multinomial Logistic Regression (MLR) classifiers where used to access the performance of individual feature sets. In the analysis performed Histogram Signature (HS) feature set combined with EMD classifier, achieves the lowest overall accuracy (80.2 %). Co-occurrence based feature set presented the highest overall classification accuracy (99.3 %) when combined with k-NN classifier. However, both Run Length and Co-occurrence based feature sets, presented robustness against classifier choice and higher classification accuracy than HS feature set.