Lilla Boroczky

Computer-aided diagnosis of pulmonary nodules using a two-step approach for feature selection and classifier ensemble construction

OBJECTIVE Accurate classification methods are critical in computer-aided diagnosis (CADx) and other clinical decision support systems. Previous research has reported on methods for combining genetic algorithm (GA) feature selection with ensemble classifier systems in an effort to increase classification accuracy. In this study, we describe a CADx system for pulmonary nodules using a two-step supervised learning system combining a GA with the random subspace method (RSM), with the aim of exploring algorithm design parameters and demonstrating improved classification performance over either the GA or RSM-based ensembles alone. METHODS AND MATERIALS We used a retrospective database of 125 pulmonary nodules (63 benign; 62 malignant) with CT volumes and clinical history. A total of 216 features were derived from the segmented image data and clinical history. Ensemble classifiers using RSM or GA-based feature selection were constructed and tested via leave-one-out validation with feature selection and classifier training executed within each iteration. We further tested a two-step approach using a GA ensemble to first assess the relevance of the features, and then using this information to control feature selection during a subsequent RSM step. The base classification was performed using linear discriminant analysis (LDA). RESULTS The RSM classifier alone achieved a maximum leave-one-out Az of 0.866 (95% confidence interval: 0.794-0.919) at a subset size of s=36 features. The GA ensemble yielded an Az of 0.851 (0.775-0.907). The proposed two-step algorithm produced a maximum Az value of 0.889 (0.823-0.936) when the GA ensemble was used to completely remove less relevant features from the second RSM step, with similar results obtained when the GA-LDA results were used to reduce but not eliminate the occurrence of certain features. After accounting for correlations in the data, the leave-one-out Az in the two-step method was significantly higher than in the RSM and the GA-LDA. CONCLUSIONS We have developed a CADx system for evaluation of pulmonary nodule based on a two-step feature selection and ensemble classifier algorithm. We have shown that by combining classifier ensemble algorithms in this two-step manner, it is possible to predict the malignancy for solitary pulmonary nodules with a performance exceeding that of either of the individual steps.

Feature Subset Selection for Improving the Performance of False Positive Reduction in Lung Nodule CAD

In this paper, we propose a feature subset selection method based on genetic algorithms to improve the performance of false positive reduction in lung nodule CAD. It is coupled with a classifier based on support vector machines. The proposed approach determines automatically the optimal size of the feature set and chooses the most relevant features from a feature pool. Its performance was tested using a lung nodule database (66 true nodules and 123 false ones) acquired by multi-slice CT scans. From 23 features calculated for each detected structure, the suggested method determined 9 as the optimal feature subset size and selected the nine features. A support vector machine-based classifier trained with the optimal feature subset has resulted in 92.4% sensitivity and 85.4% specificity using leave-one-out cross validation. Experiments also showed significant improvement achieved by a system incorporating the proposed method over a system without it. It can be also applied to other machine learning problems: e.g. computer-aided diagnosis of lung nodules.

A Two-Step Approach for Feature Selection and Classifier Ensemble Construction in Computer-Aided Diagnosis

Accurate classification methods are critical in computer-aided diagnosis and other clinical decision support systems. Previous research has studied methods for combining genetic algorithms for feature selection with ensemble classifier systems in an effort to increase classification accuracy. We propose a two-step approach that first uses genetic algorithms to reduce the number of features used to characterize the data, then applies the random subspace method on the remaining features to create a set of diverse but high performing classifiers. These classifiers are combined using ensemble learning techniques to yield a final classification. We demonstrate this approach for computer-aided diagnosis of solitary pulmonary nodules from CT scans, in which the proposed method outperforms several previously described methods.

A new enhancement method for digital video applications

This paper describes a new method to enhance the picture quality of digital video using coding information from the MPEG-2 bitstream. Experimental results show significant improvements with respect to traditional algorithms for video sequences containing coding artifacts. The proposed method can be adapted to various video enhancement algorithms (e.g. sharpness enhancement as presented in this paper) as well as to different coding standards (e.g. MPEG-4). Consequently, the proposed enhancement method can be incorporated into future digital video products (e.g. DVD, set- top boxes, digital TVs, etc) in order to achieve superior picture quality for encoded video.

Feature subset selection for improving the performance of false positive reduction in lung nodule CAD

We propose a feature subset selection method based on genetic algorithms to improve the performance of false positive reduction in lung nodule computer-aided detection (CAD). It is coupled with a classifier based on support vector machines. The proposed approach determines automatically the optimal size of the feature set, and chooses the most relevant features from a feature pool. Its performance was tested using a lung nodule database (52 true nodules and 443 false ones) acquired by multislice CT scans. From 23 features calculated for each detected structure, the suggested method determined ten to be the optimal feature subset size, and selected the most relevant ten features. A support vector machine classifier trained with the optimal feature subset resulted in 100% sensitivity and 56.4% specificity using an independent validation set. Experiments show significant improvement achieved by a system incorporating the proposed method over a system without it. This approach can be also applied to other machine learning problems; e.g. computer-aided diagnosis of lung nodules

Adaptive algorithms for pel-recursive displacement estimation

ABSTRACT In this paper adaptive algorithms for pel-recursive displacement estimation are introduced. The proposedalgorithms are similar in form to the original Wiener-based algorithm, where the extensions are the appro-priate tuning of the so-called damping parameter and the use of a linear search technique. The proposedtechniques maintain the order of complexity of the original Wiener-based displacement estimator but they are more robust and exhibit a higher rate of convergence. The adaption of the damping parameter in a Total Least Squares sense improves the performance of the estimator with only a modest increase in thecomputational complexity. The application of a bi-section linear search technique is the most effectiveextension, but the most computational intensive too. 1. INTRODUCTION Research in the field of motion estimation is concerned with the extraction of motion information cor-responding to object or camera movements from a sequence of time-varying images. The estimation ofmotion information from a sequence of images has its importance in the field of image sequence processing.Particularly in image sequence coding, motion estimation is a key issue since the coding efficiency can beimproved by exploiting the temporal correlation between the intensities in image sequences along the so-called motion trajectories [1]. In this paper, the motion is represented by a so-called motion field which isthe projection onto the image plane of 3-D object movement. The determination of each vector individuallyfrom the local time-varying intensities is an ill-posed problem since one temporal intensity variation is notsufficient to reveal the two unknown components of the corresponding motion field vector. In image cod-ing, motion estimation techniques are based mainly on such local 2-D translational motion models and twoclasses of commonly used techniques are block-matching and pel-recursive algorithms [1] .Inblock-matchingthe ill-posedness is circumvented by assigning one displacement vector to a block of pel-locations and inthe pel-recursive algorithms by using a specific recursion technique. This paper exclusively deals with thepel-recursive techniques since they provide, as opposed to block-matching, a full-resolution motion fieldwith fractional-pel accuracy.Pel-recursive techniques are based on a predictor/update form that is applied recursively along each scan-line. Research in this field has been focussing on motion detection techniques, vector prediction formulas andupdating strategies. The results presented in this paper fall in the third category. The update part mainlydetermines the rate of convergence and the estimation accuracy. Update equations presented in literature

Artifact reduction for MPEG-2 encoded video using a unified metric for digital video processing

In this paper we propose a new deringing algorithm for MPEG-2 encoded video. It is based on a Unified Metric for Digital Video Processing (UMDVP) and therefore directly linked to the coding characteristics of the decoded video. Experiments carried out on various video sequences have shown noticeable improvement in picture quality and the proposed algorithm outperforms the deringing algorithm described in the MPEG-4 video standard. Coding artifacts, particularly ringing artifacts, are especially annoying on large high-resolution displays. To prevent the enlargement and enhancement of the ringing artifacts, we have applied the proposed deringing algorithm prior to resolution enhancement. Experiments have shown that in this configuration, the new deringing algorithm has significant positive impact on picture quality.

Joint resolution enhancement and artifact reduction for MPEG-2 encoded digital video

A joint approach of resolution enhancement and artifact reduction for MPEG-2 encoded video is proposed in this paper. This new method uses a unified metric for digital video processing (UMDVP). The UMDVP is defined based on the coding information of MPEG-2 encoded video. It also takes the local scene content into account. Experimental results have demonstrated that the joint approach using the UMDVP outperforms the system of resolution enhancement and artifact reduction without the help of UMDVP. The proposed method can improve the performance of various emerging video systems using MPEG-2 compression.

Comparison of image features calculated in different dimensions for computer-aided diagnosis of lung nodules

Features calculated from different dimensions of images capture quantitative information of the lung nodules through one or multiple image slices. Previously published computer-aided diagnosis (CADx) systems have used either twodimensional (2D) or three-dimensional (3D) features, though there has been little systematic analysis of the relevance of the different dimensions and of the impact of combining different dimensions. The aim of this study is to determine the importance of combining features calculated in different dimensions. We have performed CADx experiments on 125 pulmonary nodules imaged using multi-detector row CT (MDCT). The CADx system computed 192 2D, 2.5D, and 3D image features of the lesions. Leave-one-out experiments were performed using five different combinations of features from different dimensions: 2D, 3D, 2.5D, 2D+3D, and 2D+3D+2.5D. The experiments were performed ten times for each group. Accuracy, sensitivity and specificity were used to evaluate the performance. Wilcoxon signed-rank tests were applied to compare the classification results from these five different combinations of features. Our results showed that 3D image features generate the best result compared with other combinations of features. This suggests one approach to potentially reducing the dimensionality of the CADx data space and the computational complexity of the system while maintaining diagnostic accuracy.

Comparison of computer-aided diagnosis performance and radiologist readings on the LIDC pulmonary nodule dataset

One challenge facing radiologists is the characterization of whether a pulmonary nodule detected in a CT scan is likely to be benign or malignant. We have developed an image processing and machine learning based computer-aided diagnosis (CADx) method to support such decisions by estimating the likelihood of malignancy of pulmonary nodules. The system computes 192 image features which are combined with patient age to comprise the feature pool. We constructed an ensemble of 1000 linear discriminant classifiers using 1000 feature subsets selected from the feature pool using a random subspace method. The classifiers were trained on a dataset of 125 pulmonary nodules. The individual classifier results were combined using a majority voting method to form an ensemble estimate of the likelihood of malignancy. Validation was performed on nodules in the Lung Imaging Database Consortium (LIDC) dataset for which radiologist interpretations were available. We performed calibration to reduce the differences in the internal operating points and spacing between radiologist rating and the CADx algorithm. Comparing radiologists with the CADx in assigning nodules into four malignancy categories, fair agreement was observed (κ=0.381) while binary rating yielded an agreement of (κ=0.475), suggesting that CADx can be a promising second reader in a clinical setting.