Bisser Raytchev

Computer-aided colorectal tumor classification in NBI endoscopy using local features

An early detection of colorectal cancer through colorectal endoscopy is important and widely used in hospitals as a standard medical procedure. During colonoscopy, the lesions of colorectal tumors on the colon surface are visually inspected by a Narrow Band Imaging (NBI) zoom-videoendoscope. By using the visual appearance of colorectal tumors in endoscopic images, histological diagnosis is presumed based on classification schemes for NBI magnification findings. In this paper, we report on the performance of a recognition system for classifying NBI images of colorectal tumors into three types (A, B, and C3) based on the NBI magnification findings. To deal with the problem of computer-aided classification of NBI images, we explore a local feature-based recognition method, bag-of-visual-words (BoW), and provide extensive experiments on a variety of technical aspects. The proposed prototype system, used in the experiments, consists of a bag-of-visual-words representation of local features followed by Support Vector Machine (SVM) classifiers. A number of local features are extracted by using sampling schemes such as Difference-of-Gaussians and grid sampling. In addition, in this paper we propose a new combination of local features and sampling schemes. Extensive experiments with varying the parameters for each component are carried out, for the performance of the system is usually affected by those parameters, e.g. the sampling strategy for the local features, the representation of the local feature histograms, the kernel types of the SVM classifiers, the number of classes to be considered, etc. The recognition results are compared in terms of recognition rates, precision/recall, and F-measure for different numbers of visual words. The proposed system achieves a recognition rate of 96% for 10-fold cross validation on a real dataset of 908 NBI images collected during actual colonoscopy, and 93% for a separate test dataset.

Computer-aided system for predicting the histology of colorectal tumors by using narrow-band imaging magnifying colonoscopy (with video)

BACKGROUND Narrow-band imaging (NBI) classification of colorectal lesions is clinically useful in determining treatment options for colorectal tumors. There is a learning curve, however. Accurate NBI-based diagnosis requires training and experience. In addition, objective diagnosis is necessary. Thus, we developed a computerized system to automatically classify NBI magnifying colonoscopic images. OBJECTIVE To evaluate the utility and limitations of our automated NBI classification system. DESIGN Retrospective study. SETTING Department of endoscopy, university hospital. MAIN OUTCOME MEASUREMENTS Performance of our computer-based system for classification of NBI magnifying colonoscopy images in comparison to classification by two experienced endoscopists and to histologic findings. RESULTS For the 371 colorectal lesions depicted on validation images, the computer-aided classification system yielded a detection accuracy of 97.8% (363/371); sensitivity and specificity of types B-C3 lesions for a diagnosis of neoplastic lesion were 97.8% (317/324) and 97.9% (46/47), respectively. Diagnostic concordance between the computer-aided classification system and the two experienced endoscopists was 98.7% (366/371), with no significant difference between methods. LIMITATIONS Retrospective, single-center in this initial report. CONCLUSION Our new computer-aided system is reliable for predicting the histology of colorectal tumors by using NBI magnifying colonoscopy.

Softassign and EM-ICP on GPU

In this paper we propose CUDA-based implementations of two 3D point sets registration algorithms: Soft assign and EM-ICP. Both algorithms are known for being time demanding, even on modern multi-core CPUs. Our GPUbased implementations vastly outperform CPU ones. For instance, our CUDA EM-ICP aligns 5000 points in less than 7 seconds on a GeForce 8800GT, while the same implementation in OpenMP on an Intel Core 2 Quad would take 7 minutes.

Unsupervised face recognition by associative chaining

Abstract We propose a novel method for unsupervised face recognition from time-varying sequences of face images obtained in real-world environments. The method utilizes the higher level of sensory variation contained in the input image sequences to autonomously organize the data in an incrementally built graph structure, without relying on category-specific information provided in advance. This is achieved by “chaining” together similar views across the spatio-temporal representations of the face sequences in image space by two types of connecting edges depending on local measures of similarity. Experiments with real-world data gathered over a period of several months and including both frontal and side-view faces from 17 different subjects were used to test the method, achieving correct self-organization rate of 88.6%. The proposed method can be used in video surveillance systems or for content-based information retrieval.

A computer system to be used with laser-based endoscopy for quantitative diagnosis of early gastric cancer.

Goals: To evaluate the usefulness of a newly devised computer system for use with laser-based endoscopy in differentiating between early gastric cancer, reddened lesions, and surrounding tissue. Background: Narrow-band imaging based on laser light illumination has come into recent use. We devised a support vector machine (SVM)-based analysis system to be used with the newly devised endoscopy system to quantitatively identify gastric cancer on images obtained by magnifying endoscopy with blue-laser imaging (BLI). We evaluated the usefulness of the computer system in combination with the new endoscopy system. Study: We evaluated the system as applied to 100 consecutive early gastric cancers in 95 patients examined by BLI magnification at Hiroshima University Hospital. We produced a set of images from the 100 early gastric cancers; 40 flat or slightly depressed, small, reddened lesions; and surrounding tissues, and we attempted to identify gastric cancer, reddened lesions, and surrounding tissue quantitatively. Results: The average SVM output value was 0.846±0.220 for cancerous lesions, 0.381±0.349 for reddened lesions, and 0.219±0.277 for surrounding tissue, with the SVM output value for cancerous lesions being significantly greater than that for reddened lesions or surrounding tissue. The average SVM output value for differentiated-type cancer was 0.840±0.207 and for undifferentiated-type cancer was 0.865±0.259. Conclusions: Although further development is needed, we conclude that our computer-based analysis system used with BLI will identify gastric cancers quantitatively.

A system for colorectal tumor classification in magnifying endoscopic NBI images

In this paper we propose a recognition system for classifying NBI images of colorectal tumors into three types (A, B, and C3) of structures of microvessels on the colorectal surface. These types have a strong correlation with histologic diagnosis: hyperplasias (HP), tubular adenomas (TA), and carcinomas with massive submucosal invasion (SM-m). Images are represented by Bag-of-features of the SIFT descriptors densely sampled on a grid, and then classified by an SVM with an RBF kernel. A dataset of 907 NBI images were used for experiments with 10-fold cross-validation, and recognition rate of 94.1% were obtained.

User-independent online gesture recognition by relative motion extraction

We propose a new method for user-independent gesture recognition from time-varying images. The method uses relative motion-dependent feature extraction, together with discriminant analysis and dynamically updated buffer structures for providing online learning/recognition abilities. Efficient and robust extraction/representation of information about motion is achieved. Being computationally inexpensive the method allows real-time performance.

Unsupervised face recognition from image sequences based on clustering with attraction and repulsion

We propose a new method for unsupervised face recognition from time-varying sequences of face images obtained in real-world environments. Two types of forces, attraction and repulsion, operate across the spatio-temporal facial manifolds, to autonomously organize the data without relying on any category-specific information provided in advance. Experiments with real-world data gathered over a period of several months and including both frontal and side-view faces were used to evaluate the method and encouraging results were obtained The proposed method can be used in video surveillance systems or for content-based information retrieval.

A Vision-Based Method for Recognizing Non-manual Information in Japanese Sign Language

This paper describes a vision-based method for recognizing the nonmanual information in Japanese Sign Language (JSL). This new modality information provides grammatical constraints useful for JSL word segmentation and interpretation. Our attention is focused on head motion, the most dominant non-manual information in JSL. We designed an interactive color-modeling scheme for robust face detection. Two video cameras are vertically arranged to take the frontal and profile image of the JSL user, and head motions are classified into eleven patterns. Moment-based feature and statistical motion feature are adopted to represent these motion patterns. Classification of the motion features is performed with linear discrimant analysis method. Initial experimental results show that the method has good recognition rate and can be realized in real-time.

Face sequence recognition using Grassmann Distances and Grassmann Kernels

In this paper we show how Grassmann distances and Grassmann kernels can be efficiently used to learn and classify face sequence videos. We propose two new methods, the Grassmann Distance Mutual Subspace Method (GD-MSM) which uses Grassmann distances to define the similarity between subspaces of images, and the Grassmann Kernel Support Vector Machine (GK-SVM), which applies two Grassmann kernels - the projection kernel and the Binet-Cauchy kernel - in a convex optimization scheme, using the Support Vector Machine (SVM) framework. GD-MSM and GK-SVM are compared in a face recognition task with several related methods using a large database of face image sequences from 100 subjects, containing expression changes related to a natural conversation setting. Additionally, we study the effect of combining all available training image sequences into a single subspace per category, in comparison with using multiple smaller subspaces, i.e. representing each category by several different subspaces, where each subspace is formed from image sequences taken under different conditions.