Segmentation and feature extraction techniques, with applications to MRI head studies

Exploring the knowledge contained in neuroimages: statistical discriminant analysis and automatic segmentation of the most significant changes

OBJECTIVE

The aim of this article is to propose an integrated framework for extracting and describing patterns of disorders from medical images using a combination of linear discriminant analysis and active contour models.

METHODS

A multivariate statistical methodology was first used to identify the most discriminating hyperplane separating two groups of images (from healthy controls and patients with schizophrenia) contained in the input data. After this, the present work makes explicit the differences found by the multivariate statistical method by subtracting the discriminant models of controls and patients, weighted by the pooled variance between the two groups. A variational level-set technique was used to segment clusters of these differences. We obtain a label of each anatomical change using the Talairach atlas.

RESULTS

In this work all the data was analysed simultaneously rather than assuming a priori regions of interest. As a consequence of this, by using active contour models, we were able to obtain regions of interest that were emergent from the data. The results were evaluated using, as gold standard, well-known facts about the neuroanatomical changes related to schizophrenia. Most of the items in the gold standard was covered in our result set.

CONCLUSIONS

We argue that such investigation provides a suitable framework for characterising the high complexity of magnetic resonance images in schizophrenia as the results obtained indicate a high sensitivity rate with respect to the gold standard.

Reproducibility of perfusion parameters in dynamic contrast-enhanced MRI of lung and liver tumors: effect on estimates of patient sample size in clinical trials and on individual patient responses

OBJECTIVE

Dynamic contrast-enhanced MRI (DCE-MRI) is a potentially useful noninvasive technique for assessing tissue perfusion, particularly in the context of solid tumors and targeted antiangiogenic and antivascular therapies. Our aim was to determine the reproducibility of perfusion parameters derived at DCE-MRI of tumors of the lung and liver, the most common sites of metastasis.

SUBJECTS AND METHODS

Patients with lung and liver tumors underwent two sequential DCE-MRI examinations 2-7 days apart without any intervening therapy. The volume transfer constant between blood plasma and the extravascular extracellular space (K(trans)) and blood-normalized initial area under the signal intensity-time curve (initial AUC(BN)) were computed with a two-compartment pharmacokinetic model. Differences in parameters were assessed with within-patient coefficients of variation.

RESULTS

Twenty-three patients had evaluable tumors (12 lung, 11 liver). The within-patient coefficients of variation for K(trans) and initial AUC(BN) for liver lesions were 8.9% and 9.9% and for lung lesions were 17.9% and 18.2%. Sample sizes for reductions in these parameters from 10% to 50% were estimated to range from two to 102 subjects. Estimates of confidence that changes observed in a given patient were due to intervening therapy rather than variability of the technique were calculated to range from 71% to 87% if a 20% reduction in a parameter was observed.

CONCLUSION

The rate of reproducibility of DCE-MRI parameters is in the range of 10%-20% and is influenced by lesion location, parameters being significantly more reproducible in the liver than in the lung. These findings provide the foundation for interpretation of results and design of clinical trials in which DCE-MRI studies are used to assess objective responses.

A method to fuse fMRI tasks through spatial correlations: applied to schizophrenia

Single task analysis methods of functional MRI brain data, though useful, are not able to evaluate the joint information between tasks. Data fusion of multiple tasks that probe different cognitive processes provides knowledge of the joint information and may be important in order to better understand complex disorders such as schizophrenia. In this article, we introduce a simple but effective technique to fuse two tasks by computing the histogram of correlations for all possible combinations of whole brain voxels. The approach was applied to data derived from healthy controls and patients with schizophrenia from four different tasks, auditory oddball (target), auditory oddball (novel), Sternberg working memory, and sensorimotor. It was found that in four out of six task combinations patients' intertask correlations were more positively correlated than controls', in one combination the controls showed more positive correlations and in another there was no significant difference. The robustness of this result was checked with several testing techniques. The four task combinations for which patients had more positive correlation occurred at different scanning sessions and the task combination that showed the opposite result occurred within the same scanning session. Brain regions that showed high intertask correlations were found for both groups and regions that correlated differently between the two groups were identified. The approach introduced finds interesting results and new differential features that cannot be achieved through traditional methods.

Comparison of manual tracing versus a semiautomatic radial measurement method in temporal lobe MRI volumetry for pharmacoresistant epilepsy

INTRODUCTION

The aim of this study was to test a modified radial semiautomated volumetry technique (radial divider technique, RDT) versus the manual volumetry technique (MVT) for proportionality of temporal subvolumes in 30 patients with drug-resistant temporal lobe epilepsy.

METHODS

Included in the study were 30 patients (15 female, 15 male; mean age 39.6 years) with pharmacoresistant epilepsy (mean duration 26.6 years). MRI studies were performed preoperatively on a 1.5-T scanner. All image processing steps and volume measurements were performed using ANALYZE software. The volumes of six subregions were measured bilaterally; these included the superior temporal gyrus (STG), middle + inferior temporal gyrus (MITG), fusiform gyrus (FG), parahippocampal gyrus (PHG), amygdala (AM), and hippocampus (HP). Linear regression was used to investigate the relationship between the comparable subvolumes obtained with MVT and RDT.

RESULTS

Very high correlations (R (2) >0.95) between RDT and MVT were observed for the STG + MITG and the STG + MITG + FG, but low correlations for the PHG subvolumes and the combined PHG + HP + AM subvolumes. These observations were independent of the side of the pathology and of hemisphere.

CONCLUSION

The two measurement techniques provided highly reliable proportional results. This series in a homogeneous group of TLE patients suggests that the much quicker RDT is suitable for determining the volume of temporolateral and laterobasal temporal lobe compartments, of both the affected and the non-affected side and the right and left hemisphere.

MR-based in vivo hippocampal volumetrics: 1. Review of methodologies currently employed

The advance of neuroimaging techniques has resulted in a burgeoning of studies reporting abnormalities in brain structure and function in a number of neuropsychiatric disorders. Measurement of hippocampal volume has developed as a useful tool in the study of neuropsychiatric disorders. We reviewed the literature and selected all English-language, human subject, data-driven papers on hippocampal volumetry, yielding a database of 423 records. From this database, the methodology of all original manual tracing protocols were studied. These protocols differed in a number of important factors for accurate hippocampal volume determination including magnetic field strength, the number of slices assessed and the thickness of slices, hippocampal orientation correction, volumetric correction, software used, inter-rater reliability, and anatomical boundaries of the hippocampus. The findings are discussed in relation to optimizing determination of hippocampal volume.

MR-based in vivo hippocampal volumetrics: 2. Findings in neuropsychiatric disorders

Magnetic resonance imaging (MRI) has opened a new window to the brain. Measuring hippocampal volume with MRI has provided important information about several neuropsychiatric disorders. We reviewed the literature and selected all English-language, human subject, data-driven papers on hippocampal volumetry, yielding a database of 423 records. Smaller hippocampal volumes have been reported in epilepsy, Alzheimer's disease, dementia, mild cognitive impairment, the aged, traumatic brain injury, cardiac arrest, Parkinson's disease, Huntington's disease, Cushing's disease, herpes simplex encephalitis, Turner's syndrome, Down's syndrome, survivors of low birth weight, schizophrenia, major depression, posttraumatic stress disorder, chronic alcoholism, borderline personality disorder, obsessive-compulsive disorder, and antisocial personality disorder. Significantly larger hippocampal volumes have been correlated with autism and children with fragile X syndrome. Preservation of hippocampal volume has been reported in congenital hyperplasia, children with fetal alcohol syndrome, anorexia nervosa, attention-deficit and hyperactivity disorder, bipolar disorder, and panic disorder. Possible mechanisms of hippocampal volume loss in neuropsychiatric disorders are discussed.

Volumetric analysis of the prefrontal cortex, amygdala, and hippocampus in major depression

Magnetic resonance imaging (MRI) studies in depressed subjects report smaller volumes of amygdala, hippocampus, inferior anterior cingulate, and the orbital prefrontal cortex (OPFC), components of the limbic-cortico-thalamic circuit. Major depression occurs more commonly in women, raising the possibility of an additional psychopathological process affecting women and not men. We sought to determine whether volumetric differences related to mood disorders are dependent on sex. Eight male and 10 female depressed subjects, meeting DSM III R criteria for a major depressive episode, and eight male and 10 female healthy volunteers had MRI scans on a 1.5 T GE Signa Advantage scanner. The regions of interest included amygdala, hippocampus, inferior anterior cingulate, and OPFC. In all analyses, regional volumes were normalized for total cerebral volume. Volumetric changes in the ROIs showed a significant sex by diagnosis interaction, indicating a different pattern of volumetric changes in depressed males compared with females relative to controls. Relative to sex-matched controls, the left inferior anterior cingulate was smaller in depressed males (23%) compared with depressed females (11%). Depressed females but not depressed males had smaller amygdala compared with controls (F-value = 4.946, p = 0.033). No significant volumetric differences were noted in the hippocampus or OPFC. No volumetric correlations were noted with clinical variables, depression subtypes, or a reported history of sexual or physical abuse. Sex may affect volumetric deficits in amygdala and anterior cingulate cortex in mood disorders, but no effects were found in the hippocampus or OPFC. The biology of mood disorders in females may differ in some aspects from males, and may contribute to the higher rate of depression in women.

Tissue segmentation on MR images of the brain by possibilistic clustering on a 3D wavelet representation

An algorithm for the segmentation of a single sequence of three-dimensional magnetic resonance (MR) images into cerebrospinal fluid, gray matter, and white matter classes is proposed. This new method is a possibilistic clustering algorithm using the fuzzy theory as frame and the wavelet coefficients of the voxels as features to be clustered. Fuzzy logic models the uncertainty and imprecision inherent in MR images of the brain, while the wavelet representation allows for both spatial and textural information. The procedure is fast, unsupervised, and totally independent of any statistical assumptions. The method is tested on a phantom image, then applied to normal and Alzheimer's brains, and finally compared with another classic brain tissue segmentation method, affording a relevant classification of voxels into the different tissue classes.

Automatic detection of hippocampal atrophy on magnetic resonance images

An automatic method for identifying hippocampal atrophy on magnetic resonance (MR) images obtained from patients with clinical evidence of temporal lobe epilepsy (TLE) is described. The method is based on the analysis of image intensity differences between patients and controls within a volume of interest (VOI) centred on the hippocampus. The core of the method is a fully automatic signal intensity-based inter-subject image registration technique. In particular, a global affine registration to a reference image is performed, followed by a local affine registration within the VOI. A mask produced by manual segmentation of the mean hippocampus for 30 control subjects enabled investigations to be restricted to a specified region of the VOI approximately corresponding to the hippocampus. Normal variations of hippocampal signal intensity were computed from images obtained for the 30 control subjects. The manual method of hippocampal volumetry, currently an important component of the pre-surgical evaluation of patients with clinical evidence of medically intractable TLE, is used to determine the lower 1st percentile limits of normal hippocampal volume. Hippocampi with volumes below this limit are defined as atrophic. We investigated whether the automatic method can correctly distinguish between 15 patients with significant hippocampal atrophy according to absolute volumes and a further 14 controls. ROC curves enabled evaluation of sensitivity and specificity in respect of an intensity threshold. 100% specificity is required when determining suitability of patients for neurosurgery, resulting in levels of 50% and 70% sensitivity in detecting atrophy in the right and left hippocampus, respectively. We propose that the method can be developed as an automatic screening procedure.

Intracranial volume change in craniosynostosis

OBJECT

There is still controversy regarding the optimum time to perform surgery for craniosynostosis. Some recommend surgery soon after birth and others delay until the age of 12 months. Intracranial pressure has been measured in an attempt to provide a scientific rationale, but many questions remain unanswered. To date, little attention has been given to intracranial volume and its changes during the first few years of life in children with craniosynostosis. The authors' goal was to focus on intracranial volume during this period and to compare measurements obtained in patients with craniosynostosis with measurements obtained in healthy individuals.

METHODS

Using the technique of segmentation, the intracranial volume of 84 children with various forms of craniosynostosis was measured on preoperative computerized tomography scans. The change in average volume that occurs with increasing age was calculated and compared with a model of normal intracranial volume growth. The age at presentation for children with craniosynostosis was 1 to 39 months; 76% of the patients were younger than 12 months. In eight patients in whom only one cranial expansion procedure was performed, postoperative intracranial volumes were measured as well. Several interesting observations emerged. 1) There was little difference in head growth between boys and girls with craniosynostosis during the first few months of life. After the age of 12 months, however, the difference in intracranial volume normally seen between the two genders was observed in the craniosynostosis group as well. 2) Excluding children with complex pansynostosis, who have smaller heads, children with all other types of craniosynostosis have similar head growth after the 1st year of life, with no difference between the number of and type of suture affected. Children with Apert's syndrome develop greater than normal intracranial volumes after the 1st year of life. 3) Although children with craniosynostosis are born with a smaller intracranial volume, by the age of 6 months volume has reached normal levels, and from that point on volume follows the pattern of normal head growth. 4) Children who presented after the age of 6 months and later developed recurrent craniosynostosis after initial successful treatment had a small intracranial volume at their initial presentation. 5) Of the patients whose postoperative intracranial volumes were measured, all but one had preoperative volumes at or above normal values, and their postoperative volumes were considerably higher than normal for their age. These children all followed a growth curve parallel to that of healthy children but at higher volume value. One patient with a smaller-than-normal initial intracranial volume was surgically treated at a very young age and, despite cranial expansion surgery, postoperative volume did not reach normal levels. It is postulated that this was due to the fact that the operation was performed at a time when craniosynostosis was still active.

CONCLUSIONS

The results of this study indicate that the underlying mechanism leading to craniosynostosis and constriction of head volume "exhausts" its effect during the first few months of life. Measurement of intracranial volume in clinical practice could be used to "fine tune" the optimum time for surgery. In late-presenting children, this may be useful in predicting possible recurrence.

Intracranial volume change in childhood

OBJECT

The goal of this study was to construct a model of normal changes in intracranial volume occurring throughout childhood from age 7 days to 15 years.

METHODS

Using the technique of segmentation on magnetic resonance images obtained in healthy children, intracranial volume was measured and plotted against age.

CONCLUSIONS

Intracranial volume in the first few months of life is on average 900 cm3 in males and 600 cm3 in females. By the age of 15 years, it increases up to 1500 cm3 in males and 1300 cm3 in females, increased by factors of 1.6 and 2.1, respectively. By the time the child reaches 2 years of age, intracranial volume has reached 77% (1150 cm3 in males and 1000 cm3 in females) and, by 5 years, 90% (1350 cm3 in males and 1200 cm3 in females) of the volume observed at age 15 years. The change in intracranial volume that occurs with age is not linear, but there seems to be a segmental pattern. Three main periods can be distinguished, each lasting approximately 5 years (0-5, 5-10, and 10-15 years), during which the growth of intracranial volume is linear. Throughout childhood, males have higher intracranial volumes than females, with a similar growth pattern.

Two-dimensional multi-criterion segmentation of pulmonary nodules on helical CT images

A multi-criterion algorithm for automatic delineation of small pulmonary nodules on helical CT images has been developed. In a slice-by-slice manner, the algorithm uses density, gradient strength, and a shape constraint of the nodule to automatically control segmentation process. The multiple criteria applied to separation of the nodule from its surrounding structures in lung are based on the fact that typical small pulmonary nodules on CT images have high densities, show a distinct difference in density at the boundary, and tend to be compact in shape. Prior to the segmentation, a region-of-interest containing the nodule is manually selected on the CT images. Then the segmentation process begins with a high density threshold that is decreased stepwise, resulting in expansion of the area of nodule candidates. This progressive region growing approach is terminated when subsequent thresholds provide either a diminished gradient strength of the nodule contour or significant changes of nodule shape from the compact form. The shape criterion added to the algorithm can effectively prevent the high density surrounding structures (e.g., blood vessels) from being falsely segmented as nodule, which occurs frequently when only the gradient strength criterion is applied. This has been demonstrated by examples given in the Results section. The algorithm's accuracy has been compared with that of radiologist's manual segmentation, and no statistically significant difference has been found between the nodule areas delineated by radiologist and those obtained by the multi-criterion algorithm. The improved nodule boundary allows for more accurate assessment of nodule size and hence nodule growth over a short time period, and for better characterization of nodule edges. This information is useful in determining malignancy status of a nodule at an early stage and thus provides significant guidance for further clinical management.

Segmentation of T1 MR scans for reconstruction of resistive head models

This paper describes a segmentation method primarily developed for reconstructing resistive head models for electroencephalographic modelling purposes. The method was implemented by combining several image processing techniques, such as amplitude segmentation, region growing, and image fusion. Also a graphical user interface was developed to enable semiautomatic approach to the segmentation process. This method was developed especially for segmentation of the brain and skull from T1-weighted magnetic resonance images, but can also be applied in any segmentation procedure. The entire project was implemented successfully in a PC-based computer running the Unix/NeXTstep operating system.

A novel volumetric feature extraction technique with applications to MR images

A semiautomated feature extraction algorithm is presented for the extraction and measurement of the hippocampus from volumetric magnetic resonance imaging (MRI) head scans. This algorithm makes use of elements of both deformable model and region growing techniques and allows incorporation of a priori operator knowledge of hippocampal location and shape. Experimental results indicate that the algorithm is able to estimate hippocampal volume and asymmetry with an accuracy which approaches that of laborious manual outlining techniques.

Statistical approach to segmentation of single-channel cerebral MR images

A statistical model is presented that represents the distributions of major tissue classes in single-channel magnetic resonance (MR) cerebral images. Using the model, cerebral images are segmented into gray matter, white matter, and cerebrospinal fluid (CSF). The model accounts for random noise, magnetic field inhomogeneities, and biological variations of the tissues. Intensity measurements are modeled by a finite Gaussian mixture. Smoothness and piecewise contiguous nature of the tissue regions are modeled by a three-dimensional (3-D) Markov random field (MRF). A segmentation algorithm, based on the statistical model, approximately finds the maximum a posteriori (MAP) estimation of the segmentation and estimates the model parameters from the image data. The proposed scheme for segmentation is based on the iterative conditional modes (ICM) algorithm in which measurement model parameters are estimated using local information at each site, and the prior model parameters are estimated using the segmentation after each cycle of iterations. Application of the algorithm to a sample of clinical MR brain scans, comparisons of the algorithm with other statistical methods, and a validation study with a phantom are presented. The algorithm constitutes a significant step toward a complete data driven unsupervised approach to segmentation of MR images in the presence of the random noise and intensity inhomogeneities.

Multiple resolution Bayesian segmentation of ultrasound images

We propose a novel method for obtaining the maximum a posteriori (MAP) probabilistic segmentation of speckle-laden ultrasound images. Our technique is multiple-resolution based, and relies on the conversion of speckle images with Rayleigh statistics to subsampled images with Gaussian statistics. This conversion reduces computation time, as well as allowing accurate parameter estimation for a probabilistic segmentation algorithm. Results appear to provide improvements over previous techniques in terms of low-contrast detail and accuracy.