Thomas Stehle

Texture Classification by Modeling Joint Distributions of Local Patterns With Gaussian Mixtures

Texture classification generally requires the analysis of patterns in local pixel neighborhoods. Statistically, the underlying processes are comprehensively described by their joint probability density functions (jPDFs). Even for small neighborhoods, however, stable estimation of jPDFs by joint histograms (jHSTs) is often infeasible, since the number of entries in the jHST exceeds by far the number of pixels in a typical texture region. Moreover, evaluation of distance functions between jHSTs is often computationally prohibitive. Practically, the number of entries in a jHST is therefore reduced by considering only two-pixel patterns, leading to 2D-jHSTs known as cooccurrence matrices, or by quantization of the gray levels in local patterns to only two gray levels, yielding local binary patterns (LBPs). Both approaches result in a loss of information. We introduce here a framework for supervised texture classification which reduces or avoids this information loss. Local texture neighborhoods are first filtered by a filter bank. Without further quantization, the jPDF of the filter responses is then described parametrically by Gaussian mixture models (GMMs). We show that the parameters of the GMMs can be reliably estimated from small image regions. Moreover, distances between the thus modelled jPDFs of different texture patterns can be computed efficiently in closed form from their model parameters. We furthermore extend this texture descriptor to achieve full invariance to rotation. We evaluate the framework for different filter banks on the Brodatz texture set. We first show that combining the LBP difference filters with the GMM-based density estimator outperforms the classical LBP approach and its codebook extensions. When replacing these-rather elementary-difference filters by the wavelet frame transform (WFT), the performance of the framework on all 111 Brodatz textures exceeds the one obtained more recently by spin image and RIFT descriptors by Lazebnik et al.

Classification of Colon Polyps in NBI Endoscopy using Vascularization Features

The evolution of colon cancer starts with colon polyps. There are two different types of colon polyps, namely hyperplasias and adenomas. Hyperplasias are benign polyps which are known not to evolve into cancer and, therefore, do not need to be removed. By contrast, adenomas have a strong tendency to become malignant. Therefore, they have to be removed immediately via polypectomy. For this reason, a method to differentiate reliably adenomas from hyperplasias during a preventive medical endoscopy of the colon (colonoscopy) is highly desirable. A recent study has shown that it is possible to distinguish both types of polyps visually by means of their vascularization. Adenomas exhibit a large amount of blood vessel capillaries on their surface whereas hyperplasias show only few of them. In this paper, we show the feasibility of computer-based classification of colon polyps using vascularization features. The proposed classification algorithm consists of several steps: For the critical part of vessel segmentation, we implemented and compared two segmentation algorithms. After a skeletonization of the detected blood vessel candidates, we used the results as seed points for the Fast Marching algorithm which is used to segment the whole vessel lumen. Subsequently, features are computed from this segmentation which are then used to classify the polyps. In leave-one-out tests on our polyp database (56 polyps), we achieve a correct classification rate of approximately 90%.

Local and global panoramic imaging for fluorescence bladder endoscopy

Endoscopic treatment of bladder cancer is more and more often based on photodynamic diagnostics (PDD), a specialized endoscopic technique where a narrow-band bluish illumination causes tumors to fluoresce reddish. Contrast between tumors and healthy bladder tissue is thus noticeably increased compared to white light endoscopy. A downside of PDD is the low illumination power, which requires that the distance between endoscope and bladder wall be kept low, thus resulting in a small field of view (FOV). We therefore describe an approach to combine several successive frames into a local PDD panorama, which provides a larger and sufficiently bright FOV for treatment. Furthermore, the endoscopic cancer treatment generally starts with a complete scan of the bladder to detect the tumors. For diagnosis, navigation and reporting, a global overview image of the bladder wall is often desired. While construction of such a global panorama can be based on the same algorithm as the local panorama, direct planar visualization of the sphere-shaped bladder may cause severe distortions. Apart from the global panorama computation itself, we therefore analyze these distortions, and provide an alternative visualization which is based on bladder depictions used in standard reporting forms and anatomy textbooks.

Real-time image composition of bladder mosaics in fluorescence endoscopy

Today, photodynamic diagnostics is commonly used in endoscopic intervention of the urinary bladder. Excited by a narrow band illumination, fluorescence markers enhance the visual contrast between benign and malignant tissue. Since in this modality the endoscope must be moved close to the bladder wall to provide sufficiently exposed images, the field of view (FOV) of the endoscope is very limited. This impedes the navigation and the re-identifying of multi-focal tumors for the physician. Thus, an image providing a larger FOV, composed from single images is highly desired during the intervention for surgery assistance. Since endoscopic mosaicking in real-time is still an open issue, we introduce a new feature-based image mosaicking algorithm for fluorescence endoscopy. Using a multi-threaded software design, the extraction of SURF features, the matching and the image stitching are separated in single processing threads. In an optimization step we discuss the trade-off between feature repeatability and processing time. After adjusting an optimal thread synchronization, the optimal workload of each thread results in a fast and real-time capable computation of image mosaics. On a standard hardware platform our algorithm performs within the RealTimeFrame framework with an update rate of 8.17 frames per second at full input image resolution (780×576). Providing a fast growing image with an extended FOV during the intervention, the mosaic displayed on a second monitor promises high potential for surgery assistance.

Differences in Regional Brain Volumes Two Months and One Year after Mild Traumatic Brain Injury

Conventional structural imaging is often normal after mild traumatic brain injury (mTBI). There is a need for structural neuroimaging biomarkers that facilitate detection of milder injuries, allow recovery trajectory monitoring, and identify those at risk for poor functional outcome and disability. We present a novel approach to quantifying volumes of candidate brain regions at risk for injury. Compared to controls, patients with mTBI had significantly smaller volumes in several regions including the caudate, putamen, and thalamus when assessed 2 months after injury. These differences persisted but were reduced in magnitude 1 year after injury, suggesting the possibility of normalization over time in the affected regions. More pronounced differences, however, were found in the amygdala and hippocampus, suggesting the possibility of regionally specific responses to injury.

Removal of Specular Reflections in Endoscopic Images

During an endoscopic examination, pictures from the inside of the human body are displayed on a computer monitor. Disturbing light reflections are often visible in these images. In this paper, we present an approach for removing these reflections and replacing them by an estimate obtained using a spectral deconvolution algorithm.

Polyp Segmentation in NBI Colonoscopy

Endoscopic screening of the colon (colonoscopy) is performed to prevent cancer and to support therapy. During intervention colon polyps are located, inspected and, if need be, removed by the investigator. We propose a segmentation algorithm as a part of an automatic polyp classification system for colonoscopic Narrow-Band images. Our approach includes multi-scale filtering for noise reduction, suppression of small blood vessels, and enhancement of major edges. Results of the subsequent edge detection are compared to a set of elliptic templates and evaluated. We validated our algorithm on our polyp database with images acquired during routine colonoscopic examinations. The presented results show the reliable segmentation performance of our method and its robustness to image variations.

A Comparison of Blood Vessel Features and Local Binary Patterns for Colorectal Polyp Classification

Colorectal cancer is the third leading cause of cancer deaths in the United States of America for both women and men. By means of early detection, the five year survival rate can be up to 90%. Polyps can to be grouped into three different classes: hyperplastic, adenomatous, and carcinomatous polyps. Hyperplastic polyps are benign and are not likely to develop into cancer. Adenomas, on the other hand, are known to grow into cancer (adenoma-carcinoma sequence). Carcinomas are fully developed cancers and can be easily distinguished from adenomas and hyperplastic polyps. A recent narrow band imaging (NBI) study by Tischendorf et al. has shown that hyperplastic polyps and adenomas can be discriminated by their blood vessel structure. We designed a computer-aided system for the differentiation between hyperplastic and adenomatous polyps. Our development aim is to provide the medical practitioner with an additional objective interpretation of the available image data as well as a confidence measure for the classification. We propose classification features calculated on the basis of the extracted blood vessel structure. We use the combined length of the detected blood vessels, the average perimeter of the vessels and their average gray level value. We achieve a successful classification rate of more than 90% on 102 polyps from our polyp data base. The classification results based on these features are compared to the results of Local Binary Patterns (LBP). The results indicate that the implemented features are superior to LBP.

CAMERA CALIBRATION FOR FISH-EYE LENSES IN ENDOSCOPYWITH AN APPLICATION TO 3D RECONSTRUCTION

Image analysis tasks such as 3D reconstruction from endoscopic images require compensation of geometric distortions introduced by the lens system. Appropriate camera calibration is thus necessary. Commonly used calibration algorithms rely on the well-known pinhole camera model, extended by parametric terms for radial distortions. In this paper, we demonstrate that these models are not appropriate if very strong distortions occur as is the case for endoscopic fish-eye lenses. As an alternative, we analyze a generic calibration algorithm published recently by Kannala and Brandt, which is based on more general projection equations. We show qualitatively and quantitatively that this algorithm is well suited to deal with significant distortions especially in the images rim regions. Furthermore, we demonstrate how images of a colon phantom that were corrected in such a manner can be used to obtain a 3D reconstruction

Absolute alignment of breathing states using image similarity derivatives

The fusion of information in medical imaging relies on accurate registration of the image content coming often from different sources. One of the strongest influences on the movement of organs is the patient’s respiration. It is known, that respiration status can be measured by comparing the projection images of the chest. Since the diaphragm compresses the soft tissue above, the level of similarity to a reference projection image in extremely inhaled or exhaled status gives an indication of the patient’s respiration status. If the images to be registered are generated under different conditions the similarity with a common reference image is calculated on different scales and therefore cannot be compared directly. The proposed solution uses two reference images acquired in extremely inhaled and exhaled position. By comparing the images with two references and by combining the similarity results, changes in respiration depth between acquisitions can be detected. With normal breathing, the similarity to one of the reference images increases while the similarity to the other one decreases over time or vice versa. If the patient’s respiration exceeds the respiration span of the reference images, the similarity to both reference images decreases. By using not only the similarity values but also their derivatives over time, changes in respiration depth therefore can be detected and the image fusion algorithm can act accordingly e.g. by removing images that exceed the valid respiration span.