Sebastian Hegenbart

Automated Marsh-like classification of celiac disease in children using local texture operators

Automated classification of duodenal texture patches with histological ground truth in case of pediatric celiac disease is proposed. The classical focus of classification in this context is a two-class problem: mucosa affected by celiac disease and unaffected duodenal tissue. We extend this focus and apply classification according to a modified Marsh scheme into four classes. In addition to other techniques used previously for classification of endoscopic imagery, we apply local binary pattern (LBP) operators and propose two new operator types, one of which adapts to the different properties of wavelet transform subbands. The achieved results are promising in that operators based on LBP turn out to achieve better results compared to many other texture classification techniques as used in earlier work. Specifically, the proposed wavelet-based LBP scheme achieved the best overall accuracy of all feature extraction techniques considered in the two-class case and was among the best in the four-class scheme. Results also show that a classification into four classes is feasible in principle however when compared to the two-class case we note that there is still room for improvement due to various reasons discussed.

Scale invariant texture descriptors for classifying celiac disease

Graphical abstract Highlights ► We test several approaches for the computer assisted diagnosis of celiac disease. ► Only scale invariant techniques are considered. ► The scale invariance of the approaches is explicitly assessed. ► Some of the methods improve the state of the art in detecting celiac disease. ► The approaches are distinctly less scale invariant than expected.

A scale- and orientation-adaptive extension of Local Binary Patterns for texture classification

Local Binary Patterns (LBPs) have been used in a wide range of texture classification scenarios and have proven to provide a highly discriminative feature representation. A major limitation of LBP is its sensitivity to affine transformations. In this work, we present a scale- and rotation-invariant computation of LBP. Rotation-invariance is achieved by explicit alignment of features at the extraction level, using a robust estimate of global orientation. Scale-adapted features are computed in reference to the estimated scale of an image, based on the distribution of scale normalized Laplacian responses in a scale-space representation. Intrinsic-scale-adaption is performed to compute features, independent of the intrinsic texture scale, leading to a significantly increased discriminative power for a large amount of texture classes. In a final step, the rotation- and scale-invariant features are combined in a multi-resolution representation, which improves the classification accuracy in texture classification scenarios with scaling and rotation significantly.

Impact of duodenal image capturing techniques and duodenal regions on the performance of automated diagnosis of celiac disease

Various techniques have been developed for an automated classification of endoscopic images. Besides the classical methods for endoscopic image capturing, new methods like the modified immersion technique have been devised and are in use. The impact of specific image capturing techniques for feature extraction and classification in automated diagnosis is unclear. This work applies several well tested methods for feature extraction and classification on images captured with the conventional and the modified immersion technique. We compare the classification rates and the impact on feature extraction of each specific capturing technique. We also compare the classification rates of different duodenal regions. Finally we advise an optimal combination of image capturing technique, duodenal region and feature extraction methods for automated celiac disease diagnosis.

Survey on computer aided decision support for diagnosis of celiac disease

Celiac disease (CD) is a complex autoimmune disorder in genetically predisposed individuals of all age groups triggered by the ingestion of food containing gluten. A reliable diagnosis is of high interest in view of embarking on a strict gluten-free diet, which is the CD treatment modality of first choice. The gold standard for diagnosis of CD is currently based on a histological confirmation of serology, using biopsies performed during upper endoscopy. Computer aided decision support is an emerging option in medicine and endoscopy in particular. Such systems could potentially save costs and manpower while simultaneously increasing the safety of the procedure. Research focused on computer-assisted systems in the context of automated diagnosis of CD has started in 2008. Since then, over 40 publications on the topic have appeared. In this context, data from classical flexible endoscopy as well as wireless capsule endoscopy (WCE) and confocal laser endomicrosopy (CLE) has been used. In this survey paper, we try to give a comprehensive overview of the research focused on computer-assisted diagnosis of CD.

Systematic assessment of performance prediction techniques in medical image classification: a case study on celiac disease

In the context of automated classification of medical images, many authors report a lack of available test data. Therefore techniques such as the leave-one-out cross validation or k-fold validation are used to assess how well methods will perform in practice. In case of methods based on feature subset selection, cross validation might provide bad estimations of how well the optimized technique generalizes on an independent data set. In this work, we assess how well cross validation techniques are suited to predict the outcome of a preferred setup of distinct test- and training data sets. This is accomplished by creating two distinct sets of images, used separately as training- and test-data. The experiments are conducted using a set of Local Binary Pattern based operators for feature extraction which are using histogram subset selection to improve the feature discrimination. Common problems such as the effects of over fitting data during cross validation as well as using biased image sets due to multiple images from a single patient are considered.

Do We Need Annotation Experts? A Case Study in Celiac Disease Classification

Inference of clinically-relevant findings from the visual appearance of images has become an essential part of processing pipelines for many problems in medical imaging. Typically, a sufficient amount labeled training data is assumed to be available, provided by domain experts. However, acquisition of this data is usually a time-consuming and expensive endeavor. In this work, we ask the question if, for certain problems, expert knowledge is actually required. In fact, we investigate the impact of letting non-expert volunteers annotate a database of endoscopy images which are then used to assess the absence/presence of celiac disease. Contrary to previous approaches, we are not interested in algorithms that can handle the label noise. Instead, we present compelling empirical evidence that label noise can be compensated by a sufficiently large corpus of training data, labeled by the non-experts.

A scale-adaptive extension to methods based on LBP using scale-normalized Laplacian of Gaussian extrema in scale-space

Local Binary Patterns and its derivatives have been widely used in the field of texture recognition over the last decade. A restriction of methods based on LBP is the variance in terms of signal scaling. This is mainly caused by the fixed LBP radius and the fixed support area of sampling points. In this work we present a general framework to enhance the scale-invariance of all LBP flavored methods, which can be applied to existing methods with minimal effort. Based on scale-normalized Laplacian of Gaussian extrema in scale-space, the global scale of a texture in question is estimated, combined with a confidence measure, to compute scale adapted patterns. By using the notion of intrinsic scales, textures are analyzed at appropriate LBP scales. A comprehensive experimental study shows that the scale-invariance of three different LBP based methods (LBP, LTP, Fuzzy LBP) is highly improved by the proposed extension.

On the implicit handling of varying distances and gastrointestinal regions in endoscopic video sequences with indication for celiac disease

We have shown in previous work that problems inherent in the automated diagnosis of standard gastroscopic videos, such as distortion and noise handling, can be handled implicitly, to some extent, by using a one-class support vector machine (SVM) classifier. A video sequence of a standard endoscopic procedure is characterized by rapid changes of perspective towards an inspected area causing various shots at different distances as well as non-predictable transits through gastrointestinal regions. In this work we examine to what extent a one-class support vector machine combined with features based on local binary patterns (LBP) variants can be used to implicitly handle varying camera distances to the mucosa as well as the non-predictable topographical changes during endoscopy.

Narrow band imaging versus white-light: What is best for computer-assisted diagnosis of celiac disease?

Recent developments of specialized endoscopic hardware with enhanced visualization capabilities such as narrow band imaging have been shown to improve the diagnostic accuracy in clinical practice. The current state-of-the-art in computer-assisted diagnosis of celiac disease (CD) in flexible endoscopy uses data captured under the modified immersion technique with white-light illumination. In this work, the potential benefits of the modified immersion technique using narrow band imaging for automated diagnosis is studied. We provide convincing experimental evidence that the imaging modality has a significant impact on the underlying feature distribution of general purpose image representations. Consequently, the design of systems for automated diagnosis requires the consideration of several factors in this context. We present a large experimental setup studying the most relevant factors for automated diagnosis of CD.