Multi-spectral brain tissue segmentation using automatically trained k-Nearest-Neighbor classification.

Conventional k-Nearest-Neighbor (kNN) classification, which has been successfully applied to classify brain tissue in MR data, requires training on manually labeled subjects. This manual labeling is a laborious and time-consuming procedure. In this work, a new fully automated brain tissue classification procedure is presented, in which kNN training is automated. This is achieved by non-rigidly registering the MR data with a tissue probability atlas to automatically select training samples, followed by a post-processing step to keep the most reliable samples. The accuracy of the new method was compared to rigid registration-based training and to conventional kNN-based segmentation using training on manually labeled subjects for segmenting gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) in 12 data sets. Furthermore, for all classification methods, the performance was assessed when varying the free parameters. Finally, the robustness of the fully automated procedure was evaluated on 59 subjects. The automated training method using non-rigid registration with a tissue probability atlas was significantly more accurate than rigid registration. For both automated training using non-rigid registration and for the manually trained kNN classifier, the difference with the manual labeling by observers was not significantly larger than inter-observer variability for all tissue types. From the robustness study, it was clear that, given an appropriate brain atlas and optimal parameters, our new fully automated, non-rigid registration-based method gives accurate and robust segmentation results. A similarity index was used for comparison with manually trained kNN. The similarity indices were 0.93, 0.92 and 0.92, for CSF, GM and WM, respectively. It can be concluded that our fully automated method using non-rigid registration may replace manual segmentation, and thus that automated brain tissue segmentation without laborious manual training is feasible.

Histogram-based selective deblurring to improve computed tomography imaging of calcifications.

OBJECTIVES: Computed tomography (CT) imaging of small high-density structures, eg, calcifications, is hampered by image blur. This study aims to deconvolve calcifications in the transverse and longitudinal directions while avoiding noise amplification and edge-ringing artifacts in the surrounding low-density structures. MATERIALS AND METHODS: A method referred to as histogram-based selective deblurring (HiSD) has been developed to generate a restored image by combining the low-intensity (ie, Hounsfield Units) information of the original image with the high-intensity information of the deconvolved image. HiSD is evaluated on phantom and in vitro atherosclerotic plaque CT images by comparing the original and restored images with their corresponding reference micro-CT images both qualitatively and quantitatively. RESULTS: HiSD reduces calcification blur in the transverse and longitudinal directions without introducing noise and ringing-artifacts in the surrounding tissues. Calcification area and volume measurements are significantly improved in the restored images (reducing on average overestimation by 32% and 83%, respectively). CONCLUSIONS: HiSD significantly improves CT visualization and quantification of small high-density structures imaged in vitro.

Evaluation of an improved technique for lumen path definition and lumen segmentation of atherosclerotic vessels in CT angiography.

Vessel image analysis is crucial when considering therapeutical options for (cardio-) vascular diseases. Our method, VAMPIRE (Vascular Analysis using Multiscale Paths Inferred from Ridges and Edges), involves two parts: a user defines a start- and endpoint upon which a lumen path is automatically defined, and which is used for initialization; the automatic segmentation of the vessel lumen on computed tomographic angiography (CTA) images. Both parts are based on the detection of vessel-like structures by analyzing intensity, edge, and ridge information. A multi-observer evaluation study was performed to compare VAMPIRE with a conventional method on the CTA data of 15 patients with carotid artery stenosis. In addition to the start- and endpoint, the two radiologists required on average 2.5 (SD: 1.9) additional points to define a lumen path when using the conventional method, and 0.1 (SD: 0.3) when using VAMPIRE. The segmentation results were quantitatively evaluated using Similarity Indices, which were slightly lower between VAMPIRE and the two radiologists (respectively 0.90 and 0.88) compared with the Similarity Index between the radiologists (0.92). The evaluation shows that the improved definition of a lumen path requires minimal user interaction, and that using this path as initialization leads to good automatic lumen segmentation results.

Accuracy comparison of a 16 and 64 multidetector-row computed tomography scanner to image small high-density structures.

OBJECTIVES: The accuracy in imaging small high-density structures is compared for 16 and 64 multidetector-row computed tomography (MDCT) scanners. MATERIALS AND METHODS: Phantom experiments and different quantification methods are used to establish size measurement accuracy, object signal, and image noise, for both MDCT systems. RESULTS: At similar scanning doses, image noise is larger ( approximately 55%) for the 64 MDCT compared with the 16 MDCT, leading to lower signal-to-noise ratios ( approximately 28% for objects <2 mm). Object spread in the xy-plane is similar for both systems; while it is reduced along the z-axis (by approximately 0.18 mm) for the 64 MDCT. Measurement accuracy of the 64 MDCT is not significantly higher (P < 0.05) compared with the 16 MDCT when using a relative-threshold corresponding to 50% of the object maximum attenuation value. However, when using a fixed-threshold, interscanner and interprotocol measurement differences are statistically significant (eg, volume relative errors are reduced by approximately 17% on average for the 64 MDCT). CONCLUSIONS: Measurement accuracy of the 16 and 64 MDCT scanners is not significantly different when using a 50% relative threshold. However, image noise is significantly larger for the 64 MDCT. Compared with a fixed-threshold based method, the 50% relative-threshold strongly reduces interscanner and interprotocol measurement dependency and improves accuracy.

Evaluation of an improved technique for automated center lumen line definition in cardiovascular image data.

The aim of the study was to evaluate a new method for automated definition of a center lumen line in vessels in cardiovascular image data. This method, called VAMPIRE, is based on improved detection of vessel-like structures. A multiobserver evaluation study was conducted involving 40 tracings in clinical CTA data of carotid arteries to compare VAMPIRE with an established technique. This comparison showed that VAMPIRE yields considerably more successful tracings and improved handling of stenosis, calcifications, multiple vessels, and nearby bone structures. We conclude that VAMPIRE is highly suitable for automated definition of center lumen lines in vessels in cardiovascular image data.

Positive and negative network correlations in temporal lobe epilepsy.

Temporal lobe seizures are accompanied by complex behavioral phenomena including loss of consciousness, dystonic movements and neuroendocrine changes. These phenomena may arise from extended neural networks beyond the temporal lobe. To investigate this, we imaged cerebral blood flow (CBF) changes during human temporal lobe seizures with single photon emission computed tomography (SPECT) while performing continuous video/EEG monitoring. We found that temporal lobe seizures associated with loss of consciousness produced CBF increases in the temporal lobe, followed by increases in bilateral midline subcortical structures. These changes were accompanied by marked bilateral CBF decreases in the frontal and parietal association cortex. In contrast, temporal lobe seizures in which consciousness was spared were not accompanied by these widespread CBF changes. The CBF decreases in frontal and parietal association cortex were strongly correlated with increases in midline structures such as the mediodorsal thalamus. These results suggest that impaired consciousness in temporal lobe seizures may result from focal abnormal activity in temporal and subcortical networks linked to widespread impaired function of the association cortex.

Selective frontal, parietal, and temporal networks in generalized seizures.

Are "generalized" seizures truly generalized? Generalized tonic-clonic seizures are classified as either secondarily generalized with local onset or primarily generalized, without known focal onset. In both types of generalized seizures widespread regions of the nervous system engage in abnormally synchronous and high-frequency neuronal firing. However, emerging evidence suggests that all neurons are not homogeneously involved; specific nodes within the network may be crucial for the propagation and behavioral manifestations of generalized tonic-clonic seizures. Study of human tonic-clonic seizures has been limited by problems with patient movement and variable seizure types. To circumvent these problems, we imaged generalized tonic-clonic seizures during electroconvulsive therapy, in which seizure type and timing are well controlled. (99m)Tc-hexamethylpropylene amine oxime injections during seizures provide a "snapshot" of cerebral blood flow that can be imaged by single photon emission computed tomography (SPECT) after seizure termination. Here we show that focal regions of frontal and parietal association cortex show the greatest relative signal increases. Involvement of the higher-order association cortex may explain the profound impairment of consciousness seen in generalized seizures. In addition, focal involvement of the dominant temporal lobe was associated with impaired retrograde verbal memory. Similar focal increases were also seen in imaging of spontaneous secondarily generalized tonic-clonic seizures. Relative sparing of many brain regions during both spontaneous and induced seizures suggests that specific networks may be more important than others in so-called generalized seizures.

Display of fused images: methods, interpretation, and diagnostic improvements.

The use of integrated visualization for medical images aims at assisting clinicians in the difficult task of mentally translating and integrating medical image data from multiple sources into a three-dimensional (3D) representation of the patient. This interpretation of the enormous amount and complexity of contemporary, multiparameter, and multimodal image data demands efficient methods for integrated presentation. This article reviews methods for fused display with the main focus on integration of functional with anatomical images. First, an overview of integrated two-dimensional (2D) and 3D medical image display techniques is presented, and topics related to the interpretation of the integrated images are discussed. Then we address the key issue for clinical acceptance, ie, whether these novel visualization techniques lead to diagnostic improvements. Methods for fused display appear to be powerful tools to assist the clinician in the retrieval of relevant information from multivariate medical image data. Evaluation of the different methods for fused display indicates that the diagnostic process improves, notably as concerns the anatomical localization (typically of functional processes), the registration procedure, enhancement of signal, and efficiency of information presentation (which increases speed of interpretation and comprehension). Consequently, fused display improves communication with referring specialists, increases confidence in the observations, and facilitates the intra- and intersubject comparison of a large part of the data from the different sources, thereby simplifying the extraction of additional, valuable information. In most diagnostic tasks the clinician is served best by providing several (interactive and flexible) 2D and 3D methods for fused display for a thorough assessment of the wealth of image information from multiple sources.

Comparison of statistical parametric mapping and SPECT difference imaging in patients with temporal lobe epilepsy.

PURPOSE: Statistical parametric mapping (SPM) is an image-analysis tool that assesses the statistical significance of cerebral blood flow (CBF) changes on a voxel-by-voxel basis, thereby removing the subjectivity inherent in conventional region-of-interest (ROI) analysis. Our platform of single-photon emission computed tomography (SPECT) ictal-interictal difference imaging in clinical epilepsy has been validated for localizing seizure onset. We extend the tools of SPM by further applying statistical measures for the significance of perfusion changes in individual patients to localize epileptogenic foci in patients with defined temporal lobe epilepsy by using paired scans in this preliminary study. METHODS: Twelve patients with pairs of periictal and interictal SPECT scans were analyzed in this comparison study between SPECT difference imaging and SPM difference analysis by using a reference database of paired normal healthy images. These 12 patients possessed seizure foci localized to the mesial temporal lobe as confirmed by surgical outcome and by hippocampal sclerosis on pathology. SPM was used to identify clusters of increased or decreased CBF in each patient in contrast to our control group. RESULTS: The regions having the most significant increased or decreased CBF by SPM analysis were in agreement with regions identified by conventional difference imaging and visual analysis by viewers blinded to the results of the SPM analysis. Differentiated further by time of radiopharmaceutical injection, six of seven patients injected within 100 s of seizure onset displayed hyperperfusion changes localized to the corresponding epileptogenic temporal lobe by both techniques. Among patients receiving injections after 100 s, both techniques showed primarily regions of hypoperfusion, which again were similar between these two methods. CONCLUSIONS: The results provide strong evidence supporting SPM difference analysis in assessing regions of significant CBF change from baseline in concordance with our current clinically used technique of SPECT ictal--interictal difference imaging in epilepsy patients. Difference analysis using SPM could serve as a useful diagnostic tool in the evaluation of seizure focus in temporal lobe epilepsy.