Search

In this paper, we propose a brain tumor segmentation and classification method for multi-modality magnetic resonance imaging scans. The data from multi-modal brain tumor segmentation challenge (MICCAI BraTS 2013) are utilized which are co-registered and skull-stripped, and the histogram matching is performed with a reference volume of high contrast. From the preprocessed images, the following features are then extracted: intensity, intensity differences, local neighborhood and wavelet texture. The integrated features are subsequently provided to the random forest classifier to predict five classes: background, necrosis, edema, enhancing tumor and non-enhancing tumor, and then these class labels are used to hierarchically compute three different regions (complete tumor, active tumor and enhancing tumor). We performed a leave-one-out cross-validation and achieved 88% Dice overlap for the complete tumor region, 75% for the core tumor region and 95% for enhancing tumor region, which is higher than the Dice overlap reported from MICCAI BraTS challenge.

BackgroundUltrasound images are difficult to segment because of their noisy and low contrast nature which makes it challenging to extract the important features. Typical intensity-gradient based approaches are not suitable for these low contrast images while it has been shown that the local phase based technique provides better results than intensity based methods for ultrasound images. The spatial feature extraction methods ignore the continuity in the heart cycle and may also capture spurious features. It is believed that the spurious features (noise) that are not consistent along the frames can be excluded by considering the temporal information. MethodsIn this paper, we present a local phase based 4D (3D+time) feature asymmetry (FA) measure using the monogenic signal. We have investigated the spatio-temporal feature extraction to explore the effect of adding time information in the feature extraction process. ResultsTo evaluate the impact of time dimension, the results of 4D based feature extraction are compared with the results of 3D based feature extraction which shows the favorable 4D feature extraction results when temporal resolution is good. The paper compares the band-pass filters (difference of Gaussian, Cauchy and Gaussian derivative) in terms of their feature extraction performance. Moreover, the feature extraction is further evaluated quantitatively by left ventricle segmentation using the extracted features. ConclusionsThe results demonstrate that the spatio-temporal feature extraction is promising in frames with good temporal resolution.

Background Optical mapping is a powerful research tool that is revolutionising study of cardiac electrophysiology. However, processing and analysis of optical mapping data is computationally challenging to design and implement, a difficulty further enhanced by novel camera technology providing higher temporal and spatial resolution. We have previously developed algorithms capable of objective processing and analysis of electrophysiological parameters acquired using a second-generation complementary metal-oxide semiconductor camera.1 Here we report development of improved algorithms packaged in a user-friendly graphical interface, capable of high-throughput processing and analysis of voltage and calcium optical mapping data from a wide spectrum of cameras. Functionality and processing speed is further improved through automated recognition of pacing frequency and analysis of activation and repolarisation of electrophysiological parameters at high spatiotemporal resolution (200 × 2400 pixels; sampling rate 1 kHz). Processing options allow averaging of multiple beats as well as individual beat segmentation within the whole experimental trace, thus allowing for study of dynamic changes in key parameters such as action potential duration (APD), activation times, conduction velocity (CV) and phase mapping. Methods and results We have compared how our software performs in terms of analysis and processing time with our previously published methods.1Results remain consistent between both methods (APD50=10.41±0.62 vs 10.36±0.63 ms; CV=24.3±1.68 vs 23.3±1.69 cm/s, new software vs published methods; n=5, murine atria). Substantial improvements in processing speed (up to 4 times) are achieved, compared to our previously published methods. This improvement, coupled with automatic recognition and segmentation of whole experiments by cycle length, enables analysis of frequency dependent beat-to-beat changes in APD, CV and activation times. We demonstrate dynamic beat-to-beat frequency-dependent changes in APD50 in isolated superfused murine atria over a long experimental timeframe. As expected, increased frequency reduces APD50 (3 Hz=12.56±0.08; 8.33Hz=11.78±0.29; 10 Hz=11.47±0.3; 12.5Hz=10.88±0.29 ms; n=3), however, extensive beat-to-beat analysis allows investigation of electophysiological changes over whole experimental protocols, potentially providing crucial insight into novel arrhythmic mechanisms. Conclusions We have developed high-throughput software for analysis and processing of optical mapping imaging data compatible with a wide range of optical mapping cameras/systems. Our software offers enhanced processing speeds of key electrophysiological parameters across the heart and allows beat-to-beat analysis of large computationally challenging datasets. Reference 1. Yue et al. Prog Biophys Mol Biol. 2014 Aug;115(2-3):340–8.

Ultrasound images are difficult to segment because of their noisy and low contrast nature which makes them challenging to extract the important features. Typical intensity-gradient based approaches are not suitable for these low contrast images while it has been shown that local phase based technique provides far better results than intensity based methods for ultrasound images. In this study, we adopt local phase based feature asymmetry (FA) measure using monogenic signal for 3-D and 4-D feature extraction. This paper presents our work on: (1) comparison between intensity gradient based feature detectors and phase based feature detector, (2) formulating a 4-D version of FA measure and comparing its results with 3-D FA measure, and (3) comparing quadrature pair of filters in terms of feature detection performance.

Contrast 3D echocardiography (C3DE) enhances the visual quality of ultrasound images in comparison with conventional 3D echocardiography (3DE). Though the image quality in C3DE is perceived to be improved for visual analysis, but it actually deteriorates for the purpose of automatic or semi-automatic analysis due to high speckle noise and intensity in homogeneity. Therefore, the left ventricle (LV) endocardial feature extraction and segmentation from C3DE images remains a challenging problem. To overcome this challenge, this work proposes a simple adaptive method to invert the appearance of C3DE image. In this inverted appearance, the LV cavity appears dark while the myocardium appears bright similar to 3DE image. Moreover, the resulting inverted image has high contrast and low noise appearance, yielding strong LV endocardium boundary and facilitating feature extraction for segmentation. The proposed method inverts the C3DE image by using a threshold that is determined automatically through image histogram analysis. Subsequently, the inverted image information is utilized in an edge-driven LV feature extraction and segmentation method. The segmentation results of LV boundary delineation are compared to manual LV boundary delineation and a good match is observed.

Identification of functional brain connectivity differences induced by certain neurological disorders from resting state functional MRI (rfMRI) is generally considered a difficult task. This challenging task requires the identification of discriminative neuroimaging markers. In this paper, we propose a two-stage algorithm to identify functional connectivity differences that can discriminate epileptic patients and healthy subjects. In the first stage, we determine the functional connectivity matrix between brain cortical regions for identification of potentially discriminative neuroimaging markers using a novel affinity propagation clustering method. Next, we propose a difference statistic to select the most discriminative connections between the cortical regions. Using selected connections and a support vector machine classifier, we achieve classification accuracy of 81.33% on unseen dataset. The results demonstrate that the proposed algorithm is capable of determining functional connections between brain regions which aid in discrimination of epileptic patients versus healthy subjects.

In this paper, we present a novel texture classification algorithm inspired by the self-assembling behavior of real ants when building live structures with their bodies. The proposed algorithm employs dyadic Gabor filter banks for extracting discriminant features from images containing multiple textures not known to the algorithm. The feature space is clustered using the novel ant tree clustering (ATC) algorithm based on the similarity of ants carrying the feature vectors. The results thus obtained show promise of the proposed approach.

Diagnosis and cure of colon cancer can be improved by efficiently classifying the colon tissue cells into normal and malignant classes. This paper presents the classification of hyperspectral colon tissue cells using morphological analysis of gland nuclei cells. The application of hyperspectral imaging technique in medical image analysis is a new domain for researchers. The main advantage in using hyperspectral imaging is the increased spectral resolution and detailed subpixel information. Biopsy slides with several microdots, where each microdot is from a distinct patient, are illuminated with a tuned light source and magnification is performed up to 400times. The proposed classification algorithm combines the hyperspectral imaging technique with linear discriminant analysis. Dimensionality reduction and cellular segmentation is achieved by independent component analysis (ICA) and k-means clustering. Morphological features, which describe the shape, orientation and other geometrical attributes, are next to be extracted. For classification, LDA is employed to discriminate tissue cells into normal and malignant classes. Implementation of LDA is simpler than other approaches; it saves the computational cost, while maintaining the performance. The algorithm is tested on a number of samples and its applicability is demonstrated with the help of measures such as classification accuracy rate and the area under the convex hull of ROC curves

We present a novel texture classification algorithm using 2-D discrete wavelet transform (DWT) and support vector machines (SVM). The DWT is used to generate feature images from individual wavelet subbands, and a local energy function is computed corresponding to each pixel of the feature images. This feature vector is first used for training and later on for testing the SVM classifier. The experimental setup consists of images from the Brodatz and MlT VisTeX texture databases and a combination of some images therein. The proposed method produces promising classification results for both single and multiple class texture analysis problems.

The classification of normal and malginant colon tissue cells is crucial to the diagnosis of colon cancer in humans. Given the right set of feature vectors, Support Vector Machines (SVMs) have been shown to perform reasonably well for the classification [4,13]. In this paper, we address the following question: how does the choice of a kernel function and its parameters affect the SVM classification performance in such a system? We show that the Gaussian kernel function combined with an optimal choice of parameters can produce high classification accuracy.

Segmentation is an early stage for the automated classification of tissue cells between normal and malignant types. We present an algorithm for unsupervised segmentation of images of hyperspectral human colon tissue cells into their constituent parts by exploiting the spatial relationship between these constituent parts. This is done by employing a modification of the conventional wavelet based texture analysis, on the projection of hyperspectral image data in the first principal component direction. Results show that our algorithm is comparable to other more computationally intensive methods which exploit the spectral characteristics of the hyperspectral imagery data.

Motivated by the self-assembling behavior of real ants, we present a novel algorithm for texture segmentation which is based on ant tree clustering of wavelet features. In a pattern recognition setting, wavelet features are extracted using either of the two subband filtering methods: discrete wavelet transform (DWT) or discrete wavelet packet transform (DWPT). The feature classification process is inspired by the self-assembling behavior observed in real ants where ants progressively become attached to an existing support and then successively to other attached ants thus building trees based on the similarity of feature vectors. The results thus obtained compare favorably to those of other recently published filtering based texture segmentation algorithms.