Cerebellum segmentation employing texture properties and knowledge based image processing: applied to normal adult controls and patients

ConvNets for automatic detection of polyglutamine SCAs from brain MRIs: state of the art applications

Polyglutamine spinocerebellar ataxias (polyQ SCAs) are a group of neurodegenerative diseases, clinically and genetically heterogeneous, characterized by loss of balance and motor coordination due to dysfunction of the cerebellum and its connections. The diagnosis of each type of polyQ SCA, alongside with genetic tests, includes medical images analysis, and its automation may help specialists to distinguish between each type. Convolutional neural networks (ConvNets or CNNs) have been recently used for medical image processing, with outstanding results. In this work, we present the main clinical and imaging features of polyglutamine SCAs, and the basics of CNNs. Finally, we review studies that have used this approach to automatically process brain medical images and may be applied to SCAs detection. We conclude by discussing the possible limitations and opportunities of using ConvNets for SCAs diagnose in the future.

Improving Cerebellar Segmentation with Statistical Fusion

The cerebellum is a somatotopically organized central component of the central nervous system well known to be involved with motor coordination and increasingly recognized roles in cognition and planning. Recent work in multi-atlas labeling has created methods that offer the potential for fully automated 3-D parcellation of the cerebellar lobules and vermis (which are organizationally equivalent to cortical gray matter areas). This work explores the trade offs of using different statistical fusion techniques and post hoc optimizations in two datasets with distinct imaging protocols. We offer a novel fusion technique by extending the ideas of the Selective and Iterative Method for Performance Level Estimation (SIMPLE) to a patch-based performance model. We demonstrate the effectiveness of our algorithm, Non-Local SIMPLE, for segmentation of a mixed population of healthy subjects and patients with severe cerebellar anatomy. Under the first imaging protocol, we show that Non-Local SIMPLE outperforms previous gold-standard segmentation techniques. In the second imaging protocol, we show that Non-Local SIMPLE outperforms previous gold standard techniques but is outperformed by a non-locally weighted vote with the deeper population of atlases available. This work advances the state of the art in open source cerebellar segmentation algorithms and offers the opportunity for routinely including cerebellar segmentation in magnetic resonance imaging studies that acquire whole brain T1-weighted volumes with approximately 1 mm isotropic resolution.

Automated Segmentation of Cerebellum Using Brain Mask and Partial Volume Estimation Map

While segmentation of the cerebellum is an indispensable step in many studies, its contrast is not clear because of the adjacent cerebrospinal fluid, meninges, and cerebra peduncle. Thus, various cerebellar segmentation methods, such as a deformable model or a template-based algorithm might exhibit incorrect segmentation of the venous sinuses and the cerebellar peduncle. In this study, we propose a fully automated procedure combining cerebellar tissue classification, a template-based approach, and morphological operations sequentially. The cerebellar region was defined approximately by removing the cerebral region from the brain mask. Then, the noncerebellar region was trimmed using a morphological operator and the brain-stem atlas was aligned to the individual brain to define the brain-stem area. The proposed method was validated with the well-known FreeSurfer and ITK-SNAP packages using the dice similarity index and recall and precision scores. As a result, the proposed method was significantly better than the other methods for the dice similarity index (0.93, FreeSurfer: 0.92, ITK-SNAP: 0.87) and precision (0.95, FreeSurfer: 0.90, ITK-SNAP: 0.93). Therefore, it could be said that the proposed method yielded a robust and accurate segmentation result. Moreover, additional postprocessing with the brain-stem atlas could improve its result.

An automatic cerebellum extraction method in T1-weighted brain MR images using an active contour model with a shape prior

PURPOSE

The objective of this paper was to automatically segment the cerebellum from T1-weighted human brain magnetic resonance (MR) images.

MATERIALS AND METHODS

The proposed method constructs a cerebellum template using five sets of 3-T MR imaging (MRI) data, which are used to determine the initial position and the shape prior of the cerebellum for the active contour model. Our formulation includes the active contour model with shape prior, which thereby maintains the shape of the template. The proposed active contour model is sequentially applied to sagittal-, coronal- and transverse-view images. To evaluate the proposed method, it is applied to BrainWeb data and a 3-T MRI data set and compared with FreeSurfer with respect to performance assessment metrics.

RESULTS

The segmented cerebellum was compared with the results from FreeSurfer. Using the manually segmented cerebellum as reference, we measured the average Jaccard coefficients of the proposed method, which were 0.882 and 0.885 for the BrainWeb data and 3-T MRI data set, respectively.

CONCLUSION

We presented the active contour model with shape prior for extracting the cerebellum from T1-weighted brain MR images. The proposed method yielded a robust and accurate segmentation result.

Development of identification of the central sulcus in brain magnetic resonance imaging

Magnetic resonance imaging (MRI) is useful in the quantitative evaluation of brain atrophy, because the superior contrast resolution facilitates separation of the gray and white matter. Quantitative assessment of brain atrophy has mainly been performed by manual measurement, which requires considerable time and effort to determine the brain volume. Therefore, computer-aided quantitative measurement methods for the diagnosis of brain atrophy are required. We have developed a method of segmenting the cerebrum, cerebellum-brainstem, and temporal lobe simultaneously on MR images obtained in a single sequence. It is important to measure the volume of not only these regions but also the frontal lobe in clinical use. However, for segmenting the frontal lobe, it is necessary to identify the Sylvian fissure and the central sulcus, which represent boundaries. Here, we developed a method of identifying the central sulcus from MR images obtained with a 1.5 T MRI scanner. The brain and the cerebrospinal fluid (CSF) regions were segmented using semiautomated segmentation method on MR images. The central sulcus shows an oblique line from the inside to the outside on the convexity view. The almost straight appearance of the central sulcus was used for segmentation of the central sulcus from the segmented CSF images. The central sulcus was identified with this technique in 77% of the images obtained by all sequences. This technique for identifying the central sulcus is very important not only for volumetry, but also for clinical diagnosis.

Textures in magnetic resonance images of the ischemic rat brain treated with an anti-inflammatory agent

Computer-based analysis of textures in magnetic resonance images provides a higher sensitivity to textural changes that cannot be recognized by the naked human eye. Thus, there is a better potential for identifying pathophysiological processes at an earlier stage or of a different character than even a trained radiologist can find. In the present study, the potential of texture analysis for in vivo identification of the administering effect of an anti-inflammatory drug in cerebral stroke in rats was evaluated. Twenty-seven Wistar rats underwent middle cerebral artery occlusion resulting in local ischemic brain infarct. One group of rats received alpha-melanocyte stimulating hormone (alpha-MSH) and a control group received saline only. T2-weighted images, apparent diffusion maps, and T2 maps were recorded by MR. Texture features were calculated in the T2-weighted images and correlated to the apparent diffusion coefficient (ADC) and the T2 values. From an array of tested texture features three independent features were tested further. Two of which were found to provide a significant discriminative classification between the control and the alpha-MSH groups. Furthermore, the same two texture features were significantly correlated to the ADCs. Thus, quantification of texture features can be helpful in detecting the effects of stroke therapy.

Texture analysis of multiple sclerosis: a comparative study

The difficulty of using magnetic resonance imaging (MRI) to support early diagnosis of multiple sclerosis (MS) stems from the subtle pathological changes in the central nervous system (CNS). In this study, texture analysis was performed on MR images of MS patients and normal controls and a combined set of texture features were explored in order to better discriminate tissues between MS lesions, normal appearing white matter (NAWM) and normal white matter (NWM). Features were extracted from gradient matrix, run-length (RL) matrix, gray level co-occurrence matrix (GLCM), autoregressive (AR) model and wavelet analysis, and were selected based on greatest difference between different tissue types. The results of the combined set of texture features were compared with our previous results of GLCM-based features alone. The results of this study demonstrated that (1) with the combined set of texture features, classification was perfect (100%) between MS lesions and NAWM (or NWM), less successful (88.89%) among the three tissue types and worst (58.33%) between NAWM and NWM; (2) compared with GLCM-based features, the combined set of texture features were better at discriminating MS lesions and NWM, equally good at discriminating MS lesions and NAWM and at all three tissue types, but less effective in classification between NAWM and NWM. This study suggested that texture analysis with the combined set of texture features may be equally good or more advantageous than the commonly used GLCM-based features alone in discriminating MS lesions and NWM/NAWM and in supporting early diagnosis of MS.

Volumetric texture segmentation by discriminant feature selection and multiresolution classification

In this paper, a multiresolution volumetric texture segmentation (M-VTS) algorithm is presented. The method extracts textural measurements from the Fourier domain of the data via subband filtering using an orientation pyramid (Wilson and Spann, 1988). A novel Bhattacharyya space, based on the Bhattacharyya distance, is proposed for selecting the most discriminant measurements and producing a compact feature space. An oct tree is built of the multivariate features space and a chosen level at a lower spatial resolution is first classified. The classified voxel labels are then projected to lower levels of the tree where a boundary refinement procedure is performed with a three-dimensional (3-D) equivalent of butterfly filters. The algorithm was tested with 3-D artificial data and three magnetic resonance imaging sets of human knees with encouraging results. The regions segmented from the knees correspond to anatomical structures that can be used as a starting point for other measurements such as cartilage extraction.

Reduced cerebellar volume and neurological soft signs in first-episode schizophrenia

Recent studies indicate that morphological and functional abnormalities of the cerebellum are associated with schizophrenia. Since the cerebellum is crucial for motor coordination, one may ask whether the respective changes are associated with motor dysfunction in the disease. To test these hypotheses in a clinical study, we investigated cerebellar volumes derived from volumetric magnetic resonance imaging of 37 first-episode patients with schizophrenia, schizophreniform or schizoaffective disorder and 18 healthy controls matched for age, gender and handedness. To control for potential interindividual differences in head size, intracranial volume was entered as a covariate. Neurological soft signs (NSS) were examined after remission of acute symptoms. Compared with the controls, patients had significantly smaller cerebellar volumes for both hemispheres. Furthermore, NSS in patients were inversely correlated with tissue volume of the right cerebellar hemisphere partialling for intracranial volume. No associations were detected between cerebellar volumes and psychopathological measures obtained at hospital admission when patients were in the acute psychotic state or after remission, treatment duration until remission, treatment response or prognostic factors, respectively. These findings support the hypothesis of cerebellar involvement in schizophrenia and indicate that the respective changes are associated with NSS.

[Study of automated segmentation of the cerebellum and brainstem on brain MR images]

MR imaging is an important method for diagnosing abnormalities of the brain. This paper presents an automated method to segment the cerebellum and brainstem for brain MR images. MR images were obtained from 10 normal subjects (male 4, female 6; 22-75 years old, average 31.0 years) and 15 patients with brain atrophy (male 3, female 12; 62-85 years of age, average 76.0 years). The automated method consisted of the following four steps: (1) segmentation of the brain on original images, (2) detection of an upper plane of the cerebellum using the Hough transform, (3) correction of the plane using three-dimensional (3D) information, and (4) segmentation of the cerebellum and brainstem using the plane. The results indicated that the regions obtained by the automated method were visually similar to those obtained by a manual method. The average rates of coincidence between the automated method and manual method were 83.0+/-9.0% in normal subjects and 86.4+/-3.6% in patients.

Texture analysis of medical images

The analysis of texture parameters is a useful way of increasing the information obtainable from medical images. It is an ongoing field of research, with applications ranging from the segmentation of specific anatomical structures and the detection of lesions, to differentiation between pathological and healthy tissue in different organs. Texture analysis uses radiological images obtained in routine diagnostic practice, but involves an ensemble of mathematical computations performed with the data contained within the images. In this article we clarify the principles of texture analysis and give examples of its applications, reviewing studies of the technique.

Influence of MRI acquisition protocols and image intensity normalization methods on texture classification

Texture analysis methods quantify the spatial variations in gray level values within an image and thus can provide useful information on the structures observed. However, they are sensitive to acquisition conditions due to the use of different protocols and to intra- and interscanner variations in the case of MRI. The influence was studied of two protocols and four different conditions of normalization of gray levels on the discrimination power of texture analysis methods applied to soft cheeses. Thirty-two samples of soft cheese were chosen at two different ripening periods (16 young and 16 old samples) in order to obtain two different microscopic structures of the protein gel. Proton density and T(2)-weighted MR images were acquired using a spin echo sequence on a 0.2 T scanner. Gray levels were normalized according to four methods: original gray levels, same maximum for all images, same mean for all images, and dynamics limited to micro +/- 3sigma. Regions of interest were automatically defined, and texture descriptors were then computed for the co-occurrence matrix, run length matrix, gradient matrix, autoregressive model, and wavelet transform. The features with the lowest probability of error and average correlation coefficient were selected and used for classification with 1-nearest neighbor (1-NN) classifier. The best results were obtained when using the limitation of dynamics to micro +/- 3sigma, which enhanced the differences between the two classes. The results demonstrated the influence of the normalization method and of the acquisition protocol on the effectiveness of the classification and also on the parameters selected for classification. These results indicate the need to evaluate sensitivity to MR acquisition protocols and to gray level normalization methods when texture analysis is required.

Volumetric Texture Description and Discriminant Feature Selection for MRI

This paper considers the problem of classification of Magnetic Resonance Images using 2D and 3D texture measures. Joint statistics such as co-occurrence matrices are common for analysing texture in 2D since they are simple and effective to implement. However, the computational complexity can be prohibitive especially in 3D. In this work, we develop a texture classification strategy by a sub-band filtering technique that can be extended to 3D. We further propose a feature selection technique based on the Bhattacharyya distance measure that reduces the number of features required for the classification by selecting a set of discriminant features conditioned on a set training texture samples. We describe and illustrate the methodology by quantitatively analysing a series of images: 2D synthetic phantom, 2D natural textures, and MRI of human knees.