Mithun Prasad

Multi-level classification of emphysema in HRCT lung images

Emphysema is a common chronic respiratory disorder characterised by the destruction of lung tissue. It is a progressive disease where the early stages are characterised by a diffuse appearance of small air spaces, and later stages exhibit large air spaces called bullae. A bullous region is a sharply demarcated region of emphysema. In this paper, it is shown that an automated texture-based system based on co-training is capable of achieving multiple levels of emphysema extraction in high-resolution computed tomography (HRCT) images. Co-training is a semi-supervised technique used to improve classifiers that are trained with very few labelled examples using a large pool of unseen examples over two disjoint feature sets called views. It is also shown that examples labelled by experts can be incorporated within the system in an incremental manner. The results are also compared against “density mask”, currently a standard approach used for emphysema detection in medical image analysis and other computerized techniques used for classification of emphysema in the literature. The new system can classify diffuse regions of emphysema starting from a bullous setting. The classifiers built at different iterations also appear to show an interesting correlation with different levels of emphysema, which deserves more exploration.

Multi-level Classification of Emphysema in HRCT Lung Images Using Delegated Classifiers

Emphysema is a common chronic respiratory disorder characterized by the destruction of lung tissue. It is a progressive disease where the early stages are characterized by diffuse appearance of small air spaces and later stages exhibit large air spaces called bullae. A bullous region is a sharply demarcated region of emphysema. In this paper, we show that an automated texture-based system based on delegated classifiers is capable of achieving multiple levels of emphysema extraction in High Resolution Computed Tomography (HRCT) images. The key idea of delegation is that a cautious classifier makes predictions that meet a minimum level of confidence, and delegates the difficult or uncertain predictions to a more specialized classifier. In this paper, we design a two-step scenario where a first classifier chooses the examples to classify on and delegates the more difficult examples to a second classifier. We compare this technique to well known emphysema classification techniques and ensemble methods such as bagging and boosting. Comparison of the results shows that the techniques presented here are more accurate. From a medical standpoint, the classifiers built at different iterations appear to show an interesting correlation with different levels of emphysema.

Improved Quantification and Normal Limits for Myocardial Perfusion Stress–Rest Change

We aimed to improve the quantification of myocardial perfusion stress–rest changes in myocardial perfusion SPECT (MPS) studies for the optimal automatic detection of ischemia and coronary artery disease (CAD). Methods: Rest–stress 99mTc MPS studies (997 cases; 651 consecutive cases with correlating angiography and 346 cases with less than 5% likelihood (low likelihood [LLK]) of CAD) were analyzed. Normal limits for stress–rest changes were derived from additional LLK patients (40 women, 40 men). We computed the global stress–rest change (C-SR) by integrating direct stress–rest changes for each polar map pixel. Additionally, stress–rest change and total perfusion deficit (TPD) at stress were combined in 1 variable (C-TPD) for the optimal detection of CAD. Results: The area under the receiver-operating-characteristic curve (AUC) for C-SR (0.92) was larger than that for stress TPD–rest TPD (0.88) for the identification of stenosis of 70% or more (P < 0.0001). AUC (0.94) and sensitivity (90%) for C-TPD were higher than those for stress TPD (0.91 and 83%, respectively) (P < 0.0001), whereas specificity remained the same (81%). Conclusion: C-SR and C-TPD provide higher diagnostic performance than difference between stress and rest TPD or stress hypoperfusion analysis.

Quantification of 3D regional myocardial wall thickening from gated magnetic resonance images

To develop 3D quantitative measures of regional myocardial wall motion and thickening using cardiac magnetic resonance imaging (MRI) and to validate them by comparison to standard visual scoring assessment.

Detection of bronchovascular pairs on HRCT lung images through relational learning

The identification of bronchovascular pairs on high resolution computer tomography (HRCT) images provides valuable diagnostic information in patients with suspected airway diseases. Classification of a bronchovascular pair primarily formed by two structures, namely a bronchus and a vessel, is based on relations. Therefore, classifications based on simple attributes are insufficient for the recognition of bronchovascular pairs. To address this, we make use of relations and inductive learning from examples. Relations of potential bronchovascular pairs are extracted using image analysis and used for learning within FOIL, a first order relational learning system. The system was tested on 47 images using the learned classifier and its performance was visually validated with the help of radiologists in our team.

DESIGNING RELEVANT FEATURES FOR CONTINUOUS DATA SETS USING ICA

Isolating relevant information and reducing the dimensionality of the original data set are key areas of interest in pattern recognition and machine learning. In this paper, a novel approach to reducing dimensionality of the feature space by employing independent component analysis (ICA) is introduced. While ICA is primarily a feature extraction technique, it is used here as a feature selection/construction technique in a generic way. The new technique, called feature selection based on independent component analysis (FS_ICA), efficiently builds a reduced set of features without loss in accuracy and also has a fast incremental version. When used as a first step in supervised learning, FS_ICA outperforms comparable methods in efficiency without loss of classification accuracy. For large data sets as in medical image segmentation of high-resolution computer tomography images, FS_ICA reduces dimensionality of the data set substantially and results in efficient and accurate classification.

Myocardial wall thickening from gated magnetic resonance images using Laplace's equation

The aim of our work is to present a robust 3D automated method for measuring regional myocardial thickening using cardiac magnetic resonance imaging (MRI) based on Laplaces equation. Multiple slices of the myocardium in short-axis orientation at end-diastolic and end-systolic phases were considered for this analysis. Automatically assigned 3D epicardial and endocardial boundaries were fitted to short-axis and long axis slices corrected for breathold related misregistration, and final boundaries were edited by a cardiologist if required. Myocardial thickness was quantified at the two cardiac phases by computing the distances between the myocardial boundaries over the entire volume using Laplaces equation. The distance between the surfaces was found by computing normalized gradients that form a vector field. The vector fields represent tangent vectors along field lines connecting both boundaries. 3D thickening measurements were transformed into polar map representation and 17-segment model (American Heart Association) regional thickening values were derived. The thickening results were then compared with standard 17-segment 6-point visual scoring of wall motion/wall thickening (0=normal; 5=greatest abnormality) performed by a consensus of two experienced imaging cardiologists. Preliminary results on eight subjects indicated a strong negative correlation (r=-0.8, p<0.0001) between the average thickening obtained using Laplace and the summed segmental visual scores. Additionally, quantitative ejection fraction measurements also correlated well with average thickening scores (r=0.72, p<0.0001). For segmental analysis, we obtained an overall correlation of -0.55 (p<0.0001) with higher agreement along the mid and apical regions (r=-0.6). In conclusion 3D Laplace transform can be used to quantify myocardial thickening in 3D.

Automatic Detection of Bronchial Dilatation in HRCT Lung Images

Bronchiectasis is an airway disease caused by the dilatation of the bronchial tree, and a bronchovascular pair is formed between a bronchus and a vessel. An abnormal bronchovascular pair is one that has a larger bronchus compared to its accompanying vessel. Typically, bronchi and vessels running perpendicular to the plane of section appear as near-circular rings on computed tomography (CT) scans. This paper describes BV_pairs, a system capable of detecting abnormal bronchovascular pairs in high-resolution CT scans of sparse datasets using a three-stage process: (1) detection of potential bronchovascular pairs, (2) detection of discrete pairs, where there exists no ambiguity as to the artery that accompanies a bronchus, and (3) identification of abnormal pairs with severity levels. The system was evaluated at every stage. The automated scoring for the presence and severity of bronchial abnormalities was demonstrated to be comparable to that of an experienced radiologist (i.e., kappa statistics κ > 0.5). In addition, BV_pairs was also evaluated on images containing honeycombing regions, since honeycombing cysts appear very similar to bronchi, and the system could successfully differentiate honeycombing cysts from bronchi.

Multi-view learning for bronchovascular pair detection

In many important image classification problems, acquiring class labels for training instances is costly, while gathering large quantities of unlabelled data is cheap. A semiautomated system for the classification of bronchovascular pairs based on co-training in high resolution computed tomography (HRCT) images is presented. A bronchovascular pair is formed between a bronchus and a vessel. The identification of such structures provides valuable diagnostic information in patients with suspected airway diseases. Co-training is a semi-supervised multi-view learning algorithm where classifiers trained with a small number of labelled examples are improved by augmenting the small training set with a large pool of unseen examples. We incorporate active learning where the user labels examples on which the two views disagree. The two views in our system are based on spatial relations and ERS, a gradient based feature set. In addition, the optimal parameters required in the pre-processing step before feature extraction and recognition was automatically chosen. The system was co-trained on 41 unlabelled HRCT scans selected from 26 patient studies. It was successfully evaluated on 26 other HRCT scans manually labelled in consultation with radiologists.

Automatic Detection of Tram Tracks on HRCT Images

On high resolution computed tomography (HRCT) images, dilated airways appear as two parallel lines that resemble tram tracks, when they lie in the plane of scan. Tram tracks, when visible, are characteristic of bronchiectasis, a disease caused by the irreversible dilatation of the bronchial tree. Detection of such patterns provides valuable diagnostic information. In this work, semi-supervised learning together with image analysis techniques have been used to detect tram tracks on HRCT images. The approach was tested on 1091 HRCT images belonging to 54 patients, and the results visually validated by radiologists. Sensitivity and specificity of 80% and 91% respectively were achieved.