Automatic tumor segmentation using knowledge-based techniques

A wavelet-based optimal texture feature set for classification of brain tumours

In this work, the classification of brain tumours in magnetic resonance images is studied by using optimal texture features. These features are used to classify three sets of brain images - normal brain, benign tumour and malignant tumour. A wavelet-based texture feature set is derived from the region of interest. Each selected brain region of interest is characterized with both its energy and texture features extracted from the selected high frequency subband. An artificial neural network classifier is employed to evaluate the performance of these features. Feature selection is performed by a genetic algorithm. Principal component analysis and classical sequential methods are compared against the genetic approach in terms of the best recognition rate achieved and the optimal number of features. A classification performance of 98% is achieved in a genetic algorithm with only four of the available 29 features. Principal component analysis and classical sequential methods require a larger feature set to attain the similar classification accuracy of 98%. The optimal texture features such as range of angular second moment, range of sum variance, range of information measure of correlation II and energy selected by the genetic algorithm provide best classification performance with lower computational effort.

Asymmetry analysis in rodent cerebral ischemia models

RATIONALE AND OBJECTIVES

An automated method for identification and segmentation of acute/subacute ischemic stroke, using the inherent bi-fold symmetry in brain images, is presented. An accurate and automated method for localization of acute ischemic stroke could provide physicians with a mechanism for early detection and potentially faster delivery of effective stroke therapy.

MATERIALS AND METHODS

Segmentation of ischemic stroke was performed on magnetic resonance (MR) images of subacute rodent cerebral ischemia. Eight adult male Wistar rats weighing 225-300 g were anesthetized with halothane in a mix of 70% nitrous oxide/30% oxygen. Animal core temperature was maintained at 37 degrees C during the entire surgical procedure, including occlusion of the middle cerebral artery (MCA) and the 90-minute post-reperfusion period. To confirm cerebral ischemia, transcranial measurements of cerebral blood flow were performed with laser-Doppler flowmetry, using 15-mm flexible fiberoptic Doppler probes attached to the skull over the MCA territory. Animal MR scans were performed at 1.5 T using a knee coil. Three experts performed manual tracing of the stroke regions for each rat, using the histologic-stained slices to guide delineation of stroke regions. A strict tracing protocol was followed that included multiple (three) tracings of each stroke region. The volumetric MR image data were processed for each rat by computing the axis of symmetry and extracting statistical dissimilarities. A nonparametric Wilcoxon rank sum test operating on paired windows in opposing hemispheres identified seeds in the pixels exhibiting statistically significant bi-fold mirror asymmetry. Two brain reference maps were used for analysis: an absolute difference map (ADM) and a statistical difference map (SDM). Although an ADM simply displays the absolute difference by subtracting one brain hemisphere from its reflection, SDM highlights regions by labeling pixels exhibiting statistically significant asymmetry.

RESULTS

To assess the accuracy of the proposed segmentation method, the surrogate ground truth (the stroke tracing data) was compared to the results of our proposed automated segmentation algorithm. Three accuracy segmentation metrics were utilized: true-positive volume fraction (TPVF), false-positive volume fraction (FPVF), and false-negative volume fraction (FNVF). The mean value of the TPVF for our segmentation method was 0.8877; 95% CI 0.7254 to 1.0500; the mean FPVF was 0.3370, 95% CI -0.0893 to 0.7633; the mean FNVF was 0.1122, 95% CI -0.0502 to 0.2747.

CONCLUSIONS

Unlike most segmentation methods that require some degree of manual intervention, our segmentation algorithm is fully automated and highly accurate in identifying regions of brain asymmetry. This approach is attractive for numerous neurologic applications where the operator's intervention should be minimal or null.

Multiparametric tissue characterization of brain neoplasms and their recurrence using pattern classification of MR images

RATIONALE AND OBJECTIVES

Treatment of brain neoplasms can greatly benefit from better delineation of bulk neoplasm boundary and the extent and degree of more subtle neoplastic infiltration. Magnetic resonance imaging (MRI) is the primary imaging modality for evaluation before and after therapy, typically combining conventional sequences with more advanced techniques such as perfusion-weighted imaging and diffusion tensor imaging (DTI). The purpose of this study is to quantify the multiparametric imaging profile of neoplasms by integrating structural MRI and DTI via statistical image analysis methods to potentially capture complex and subtle tissue characteristics that are not obvious from any individual image or parameter.

MATERIALS AND METHODS

Five structural MRI sequences, namely, B0, diffusion-weighted images, fluid-attenuated inversion recovery, T1-weighted, and gadolinium-enhanced T1-weighted, and two scalar maps computed from DTI (ie, fractional anisotropy and apparent diffusion coefficient) are used to create an intensity-based tissue profile. This is incorporated into a nonlinear pattern classification technique to create a multiparametric probabilistic tissue characterization, which is applied to data from 14 patients with newly diagnosed primary high-grade neoplasms who have not received any therapy before imaging.

RESULTS

Preliminary results demonstrate that this multiparametric tissue characterization helps to better differentiate among neoplasm, edema, and healthy tissue, and to identify tissue that is likely to progress to neoplasm in the future. This has been validated on expert assessed tissue.

CONCLUSION

This approach has potential applications in treatment, aiding computer-assisted surgery by determining the spatial distributions of healthy and neoplastic tissue, as well as in identifying tissue that is relatively more prone to tumor recurrence.

Efficient multilevel brain tumor segmentation with integrated bayesian model classification

We present a new method for automatic segmentation of heterogeneous image data that takes a step toward bridging the gap between bottom-up affinity-based segmentation methods and top-down generative model based approaches. The main contribution of the paper is a Bayesian formulation for incorporating soft model assignments into the calculation of affinities, which are conventionally model free. We integrate the resulting model-aware affinities into the multilevel segmentation by weighted aggregation algorithm, and apply the technique to the task of detecting and segmenting brain tumor and edema in multichannel magnetic resonance (MR) volumes. The computationally efficient method runs orders of magnitude faster than current state-of-the-art techniques giving comparable or improved results. Our quantitative results indicate the benefit of incorporating model-aware affinities into the segmentation process for the difficult case of glioblastoma multiforme brain tumor.

Multimodal MRI segmentation of ischemic stroke lesions

The problem addressed in this paper is the automatic segmentation of stroke lesions on MR multi-sequences. Lesions enhance differently depending on the MR modality and there is an obvious gain in trying to account for various sources of information in a single procedure. To this aim, we propose a multimodal Markov random field model which includes all MR modalities simultaneously. The results of the multimodal method proposed are compared with those obtained with a mono-dimensional segmentation applied on each MRI sequence separately. We constructed an Atlas of blood supply territories to help clinicians in the determination of stroke subtypes and potential functional deficit.

Detection and segmentation of pathological structures by the extended graph-shifts algorithm

We propose an extended graph-shifts algorithm for image segmentation and labeling. This algorithm performs energy minimization by manipulating a dynamic hierarchical representation of the image. It consists of a set of moves occurring at different levels of the hierarchy where the types of move, and the level of the hierarchy, are chosen automatically so as to maximally decrease the energy. Extended graph-shifts can be applied to a broad range of problems in medical imaging. In this paper, we apply extended graph-shifts to the detection of pathological brain structures: (i) segmentation of brain tumors, and (ii) detection of multiple sclerosis lesions. The energy terms in these tasks are learned from training data by statistical learning algorithms. We demonstrate accurate results, precision and recall in the order of 93%, and also show that the algorithm is computationally efficient, segmenting a full 3D volume in about one minute.

Multispectral tissue characterization in a RIF-1 tumor model: monitoring the ADC and T2 responses to single-dose radiotherapy. Part II

A multispectral (MS) approach that combines apparent diffusion coefficient (ADC) and T(2) parameter maps with k-means (KM) clustering was employed to distinguish multiple compartments within viable tumor tissue (V1 and V2) and necrosis (N1 and N2) following single-dose (1000 cGy) radiotherapy in a radiation-induced fibrosarcoma (RIF-1) tumor model. The contributions of cell kill and tumor growth kinetics to the radiotherapy-induced response were investigated. A larger pretreatment V1 volume was correlated with decreased tumor growth delay (TGD) (r = 0.68) and cell kill (r = 0.71). There was no correlation for the pretreatment V2 volume. These results suggest that V1 tissue is well oxygenated and radiosensitive, whereas V2 tissue is hypoxic and therefore radioresistant. The relationship between an early ADC response and vasogenic edema and formation of necrosis was investigated. A trend for increased ADC was observed prior to an increase in the necrotic fraction (NF). Because there were no changes in T(2), these observations suggest that the early increase in ADC is more likely based on a slight reduction in cell density, rather than radiation-induced vasogenic edema. Quantitative assessments of individual tissue regions, tumor growth kinetics, and cell kill should provide a more accurate means of monitoring therapy in preclinical animal models because such assessments can minimize the issue of intertumor variability.

Multispectral quantification of tissue types in a RIF-1 tumor model with histological validation. Part I

Accurate assessments of therapeutic efficacy are confounded by intra- and intertumor heterogeneity. To address this issue we employed multispectral (MS) analysis using the apparent diffusion coefficient (ADC), T(2), proton density (M(0)), and k-means (KM) clustering algorithm to identify multiple compartments within both viable and necrotic tissue in a radiation-induced fibrosarcoma (RIF-1) tumor model receiving single-dose (1000 cGy) radiotherapy. Optimization of the KM method was achieved through histological validation by hematoxylin-eosin (H& and E) staining and hypoxia-inducible factor-1alpha (HIF-1alpha) immunohistochemistry. The optimum KM method was determined to be a two-feature (ADC, T(2)) and four-cluster (two clusters each of viable tissue and necrosis) segmentation. KM volume estimates for both viable (r = 0.94, P < 0.01) and necrotic (r = 0.69, P = 0.07) tissue were highly correlated with their H&E counterparts. HIF-1alpha immunohistochemistry showed that the intensity of HIF-1alpha expression tended to be concentrated in perinecrotic regions, supporting the subdivision of the viable tissue into well-oxygenated and hypoxic regions. Since both necrosis and hypoxia have been implicated in poor treatment response and reduced patient survival, the ability to quantify the degree of necrosis and the severity of hypoxia with this method may aid in the planning and modification of treatment regimens.

Extraction of tongue carcinoma using genetic algorithm-induced fuzzy clustering and artificial neural network from MR images

A novel hierarchical image segmentation approach has been developed for the extraction of tongue carcinoma from magnetic resonance (MR) images. First, a genetic algorithm (GA)-induced fuzzy clustering is used for initial segmentation of MR images of head and neck. Then these segmented masses are refined to reduce the false-positives using an artificial neural network (ANN)-based symmetry detection and image analysis procedure. The proposed technique is applied to clinical MR images of tongue carcinoma and quantitative evaluations are performed. Experimental results suggest that the proposed approach provides an effective method for tongue carcinoma extraction with high accuracy and minimal user-dependency.

Segmentation and grading of brain tumors on apparent diffusion coefficient images using self-organizing maps

An accurate computer-assisted method to perform segmentation of brain tumor on apparent diffusion coefficient (ADC) images and evaluate its grade (malignancy state) has been designed using a mixture of unsupervised artificial neural networks (ANN) and hierarchical multiresolution wavelet. Firstly, the ADC images are decomposed by multiresolution wavelets, which are subsequently selectively reconstructed to form wavelet filtered images. These wavelet filtered images along with FLAIR and T2 weighted images have been utilized as the features to unsupervised neural network - self organizing maps (SOM) - to segment the tumor, edema, necrosis, CSF and normal tissue and grade the malignant state of the tumor. A novel segmentation algorithm based on the number of hits experienced by Best Matching Units (BMU) on SOM maps is proposed. The results shows that the SOM performs well in differentiating the tumor, edema, necrosis, CSF and normal tissue pattern vectors on ADC images. Using the trained SOM and proposed segmentation algorithm, we are able to identify high or low grade tumor, edema, necrosis, CSF and normal tissue. The results are validated against manually segmented images and sensitivity and the specificity are observed to be 0.86 and 0.93, respectively.

Fuzzy modelling of different tumorous cerebral tissues on MRI images based on fusion of feature information

A fuzzy modelling approach is being proposed in this paper to estimate the model of tumorous cerebral tissues on MRI images. According to the graduality of description of neuro-radiologists, two tables have been combined to address two types of potential categories of glioma characteristics one is the array of different tissues versus gray level, and the other is the possibility of different tissues belonging to tumor. A hierarchical estimation structure has been proposed to estimate the models of tumorous cerebral tissues on MRI images by the fusion of this a priori knowledge. Through the model outline drawing, adjusting and parameters estimation, the result has shown that this is an efficient modelling method.

Physiologic characterisation of glioblastoma multiforme using MRI-based hypoxia mapping, chemical shift imaging, perfusion and diffusion maps

PURPOSE

A multiparametric, physiologic MRI approach was considered to more completely characterise biopsy-confirmed glioblastoma multiforme (GBM). Chemical shift imaging (CSI) supplied biochemical information in metabolite ratios, while perfusion images provided data on presumed vascularity from regional cerebral blood volume (rCBV) and permeability maps. Diffusion-weighted images were reduced to apparent diffusion coefficient (ADC) maps to evaluate cellularity, and blood oxygen level-dependent imaging was used to create maps of putative hypoxic regions.

METHODS

Six post-treatment GBM patients were scanned at 3-month intervals until recurrence was suggested by conventional MRI parameters, yielding 20 scans for consideration. The percentage of extreme values in each technique that overlapped with other parameters was measured and compared across hemispheres to assess utility.

RESULTS

We found significantly better performance in selecting the diseased hemisphere for overall percent overlap when compared to voxel counts from individual thresholded parameter maps. Parameters were selected on the basis of highest overlap, and corresponding composite overlap maps show increased specificity to likely recurrent regions by reducing the number of falsely positive voxels, and offer insight into relationships between various parameters.

CONCLUSION

In a pilot group of patients, percent overlap appears to be sensitive to recurrent disease. When used to combine multiple parameters, voxels containing overlap can specifically target probable recurrent areas.

Technical aspects and evaluation methodology for the application of two automated brain MRI tumor segmentation methods in radiation therapy planning

The purpose of this study was to design the steps necessary to create a tumor volume outline from the results of two automated multispectral magnetic resonance imaging segmentation methods and integrate these contours into radiation therapy treatment planning. Algorithms were developed to create a closed, smooth contour that encompassed the tumor pixels resulting from two automated segmentation methods: k-nearest neighbors and knowledge guided. These included an automatic three-dimensional (3D) expansion of the results to compensate for their undersegmentation and match the extended contouring technique used in practice by radiation oncologists. Each resulting radiation treatment plan generated from the automated segmentation and from the outlining by two radiation oncologists for 11 brain tumor patients was compared against the volume and treatment plan from an expert radiation oncologist who served as the control. As part of this analysis, a quantitative and qualitative evaluation mechanism was developed to aid in this comparison. It was found that the expert physician reference volume was irradiated within the same level of conformity when using the plans generated from the contours of the segmentation methods. In addition, any uncertainty in the identification of the actual gross tumor volume by the segmentation methods, as identified by previous research into this area, had small effects when used to generate 3D radiation therapy treatment planning due to the averaging process in the generation of margins used in defining a planning target volume.

Correlation between MR imaging–derived nasopharyngeal carcinoma tumor volume and TNM system

PURPOSE

To measure nasopharyngeal carcinoma tumor volume based on magnetic resonance images using a validated semiautomated measurement methodology and correlate tumor volume with TNM T classification.

METHODS AND MATERIALS

The study population consisted of 206 consecutive nasopharyngeal carcinoma patients who had magnetic resonance imaging staging scans. Tumor volume was measured using a semisupervised knowledge-based fuzzy clustering algorithm. Patients were divided into 4 groups according to TNM T classification. The difference in tumor volumes among the various TNM T-classification groups was examined.

RESULTS

The mean tumor volume in each T-classification group is as follows: T1, 8.6 mL +/- 5.0 (standard deviation [SD]); T2, 18.1 mL +/- 8.1 (SD); T3, 25.8 mL +/- 14.1 (SD); and T4, 36.2 mL +/- 18.9 (SD). The mean tumor volume increased significantly with advancing T classification (p < 0.0001). Tumor volume in a more advanced T group was significantly larger than that in an adjacent early T group (p < 0.01).

CONCLUSION

Validated magnetic resonance imaging-based tumor volume shows positive correlation between tumor volume and advancing T-classification groups. It may be possible to incorporate tumor volume as an additional prognostic factor into the existing TNM system.

Target Definition in Prostate, Head, and Neck

Target definition is a major source of errors in both prostate and head and neck external-beam radiation treatment. Delineation errors remain constant during the course of radiation and therefore have a large impact on the dose to the tumor. Major sources of delineation variation are visibility of the target including its extensions, disagreement on the target extension, and interpretation or lack of delineation protocols. The visibility of the target can be greatly improved with the use of multimodality imaging. Both in the head and neck and the prostate, computed tomography (CT)-magnetic resonance imaging coregistration decreases the target volume and its variability. CT-positron emission tomography delineation is promising for delineation in head and neck cancer. Despite the better visibility, a different interpretation of the target extension remains a major source of error. The use of coregistration of CT with a second modality, together with improved guidelines for delineation and an online anatomical atlas, increases agreement between observers in prostate, lung, and nasopharynx tumors. Delineation errors should not be treated differently from other geometrical errors. Similar margin recipes for the correction of setup errors and organ motion should be adapted to incorporate the effect of delineation errors. A calculation of a 3-dimensional clinical target volume-planning target volume margin incorporating delineation errors for the head and neck is around 6.1 to 9.7 mm. Given the good local control of IMRT with smaller margins and smaller pathological specimens, it is likely that the delineated CTV frequently overestimates the actual volume.

Segmentation of nasopharyngeal carcinoma (NPC) lesions in MR images

PURPOSE

An accurate and reproducible method to delineate tumor margins from uninvolved tissues is of vital importance in guiding radiation therapy (RT). In nasopharyngeal carcinoma (NPC), tumor margin may be difficult to identify in magnetic resonance (MR) images, making the task of optimizing RT treatment more difficult. Our aim in this study is to develop a semiautomatic image segmentation method for NPC that requires minimal human intervention and is capable of delineating tumor margins with good accuracy and reproducibility.

METHODS AND MATERIALS

The segmentation algorithm includes 5 stages: masking, Bayesian probability calculation, smoothing, thresholding and seed growing, and finally dilation and overlaying of results with different thresholds. The algorithm is based on information obtained from the contrast enhancement ratio of T1-weighted images and signal intensity of T2-weighted images. The algorithm is initiated by the selection of a valid anatomical seed point within the tumor by the user. The algorithm was evaluated on MR images from 7 NPC patients and was compared against the radiologist's reference outline.

RESULTS

The algorithm was successfully implemented on all 7 subjects. With a threshold of 1, the average percent match is 78.5 +/- 3.86 (standard deviation) %, and the correspondence ratio is 66.5 +/- 7%.

DISCUSSION

The segmentation algorithm presented here may be useful for diagnosing NPC and may guide RT treatment planning. Further improvement will be desirable to improve the accuracy and versatility of the method.

A system for brain tumor volume estimation via MR imaging and fuzzy connectedness

This paper presents a method for the precise, accurate and efficient quantification of brain tumor (glioblastomas) via MRI that can be used routinely in the clinic. Tumor volume is considered useful in evaluating disease progression and response to therapy, and in assessing the need for changes in treatment plans. We use multiple MRI protocols including FLAIR, T1, and T1 with Gd enhancement to gather information about different aspects of the tumor and its vicinity. These include enhancing tissue, nonenhancing tumor, edema, and combinations of edema and tumor. We have adapted the fuzzy connectedness framework for tumor segmentation in this work and the method requires only limited user interaction in routine clinical use. The system has been tested for its precision, accuracy, and efficiency, utilizing 10 patient studies. The percent coefficient of variation (% CV) in volume due to operator subjectivity in specifying seeds for fuzzy connectedness segmentation is less than 1%. The mean operator and computer time required per study for estimating the volumes of both edema and enhancing tumor is about 16 min. The software package is designed to run under operator supervision. Delineation has been found to agree with the operators' visual inspection most of the time except in some cases when the tumor is close to the boundary of the brain. In the latter case, the scalp, surgical scar, or orbital contents are included in the delineation, and an operator has to exclude this manually. The methodology is rapid, robust, consistent, yielding highly reproducible measurements, and is likely to become part of the routine evaluation of brain tumor patients in our health system.

Segmentation of tumors in magnetic resonance brain images using an interactive multiscale watershed algorithm

RATIONALE AND OBJECTIVE

This article presents the evaluation of an interactive multiscale watershed segmentation algorithm for segmenting tumors in magnetic resonance brain images of patients scheduled for neuronavigational procedures.

MATERIALS AND METHODS

The watershed method is compared with manual delineation with respect to accuracy, repeatability, and efficiency.

RESULTS

In the 20 patients included in this study, the measured volume of the tumors ranged from 2.7 to 81.9 cm(3). A comparison of the tumor volumes measured with watershed segmentation to the volumes measured with manual delineation shows that the two methods are interchangeable according to the Bland and Altman criterion, and thus equally accurate. The repeatability of the watershed method and the manual method are compared by looking at the similarity of the segmented volumes. The similarity for intraobserver and interobserver variability for watershed segmentation is 96.4% and 95.3%, respectively, compared with 93.5% and 90.0% for manual outlining, from which it may be concluded that the watershed method is more repeatable. Moreover, the watershed algorithm is on average three times faster than manual outlining.

CONCLUSION

The watershed method has an accuracy comparable to that of manual delineation and outperforms manual outlining on the criteria of repeatability and efficiency.

Nasopharyngeal Carcinoma Tumor Volume Measurement1

Tumor volume was measured in 69 patients with nasopharyngeal carcinoma. On transverse nonenhanced T1-weighted and gadolinium-enhanced T1-weighted magnetic resonance (MR) images, segmentation was performed by means of seed growing and knowledge-based fuzzy clustering methods. Data were compared with those collected with the manual tracing method and analyzed for interoperator variance and interobserver reliability. There was no significant difference between the volumes determined with manual tracing or semiautomated segmentation (P >.05). On the volume level, Pearson correction coefficients were close for both the manual tracing and semiautomated methods. Significant differences in interoperator variance existed between the two methods on the pixel level (P <.05). Compared with manual tracing, the semiautomated method helped reduce interoperator variance and obtain higher interobserver reliability. Findings in the current study validate the use of semiautomated volume measurement methods for nasopharyngeal carcinoma.

Tongue carcinoma: tumor volume measurement

PURPOSE

To validate the semiautomated methods of tongue carcinoma tumor volume measurement by comparing the conventional manual trace method with 2 semiautomated computer methods: seed growing and region deformation.

MATERIALS AND METHODS

The study population consisted of 16 patients with histology-proven tongue carcinoma. Two head-and-neck radiologists independently measured the tumor volume demonstrated on pretreatment T2-weighted magnetic resonance data sets. The tumor volumes were measured using manual tracing and semiautomated seed growing and region deformation algorithm. Data were recorded for analysis of interoperator variance and interobserver reliability at volume and pixel levels.

RESULTS

There was no significant difference between the manually traced volume and semiautomated segmentation volumes for both operators. No significant difference was found in interobserver variance among the 3 methods at volume level. However, there was significant difference between manual tracing and semiautomated segmentation methods in interobserver reliability at pixel level.

CONCLUSION

The semiautomated methods could achieve satisfactory segmentation results. They could also reduce interoperator variance and obtain a higher interobserver reliability. This study validates the use of semiautomated volume measurement methods for tongue carcinoma.

Brain tumor target volume determination for radiation treatment planning through automated MRI segmentation

PURPOSE

To assess the effectiveness of two automated magnetic resonance imaging (MRI) segmentation methods in determining the gross tumor volume (GTV) of brain tumors for use in radiation therapy treatment planning.

METHODS AND MATERIALS

Two automated MRI tumor segmentation methods (supervised k-nearest neighbors [kNN] and automatic knowledge-guided [KG]) were evaluated for their potential as "cyber colleagues." This required an initial determination of the accuracy and variability of radiation oncologists engaged in the manual definition of the GTV in MRI registered with computed tomography images for 11 glioma patients. Three sets of contours were defined for each of these patients by three radiation oncologists. These outlines were compared directly to establish inter- and intraoperator variability among the radiation oncologists. A novel, probabilistic measurement of accuracy was introduced to compare the level of agreement among the automated MRI segmentations. The accuracy was determined by comparing the volumes obtained by the automated segmentation methods with the weighted average volumes prepared by the radiation oncologists.

RESULTS

Intra- and inter-operator variability in outlining was found to be an average of 20% +/- 15% and 28% +/- 12%, respectively. Lowest intraoperator variability was found for the physician who spent the most time producing the contours. The average accuracy of the kNN segmentation method was 56% +/- 6% for all 11 cases, whereas that of the KG method was 52% +/- 7% for 7 of the 11 cases when compared with the physician contours. For the areas of the contours where the oncologists were in substantial agreement (i.e., the center of the tumor volume), the accuracy of kNN and KG was 75% and 72%, respectively. The automated segmentation methods were found to be least accurate in outlining at the edges of the tumor volume.

CONCLUSIONS

The kNN method was able to segment all cases, whereas the KG method was limited to enhancing tumors and gliomas with clear enhancing edges and no cystic formation. Both methods undersegment the tumor volume when compared with the radiation oncologists and performed within the variability of the contouring performed by experienced radiation oncologists based on the same data.

Quantification of tumour response to radiotherapy

In 1979, the World Health Organization (WHO) established criteria based on tumour volume change for classifying response to therapy as (i) progressive disease (PD), (ii) partial recovery (PR), and (iii) no change (NC). Typically, the tumour volume is reported from diameter measurements, using the calliper method. Alternatively, the Cavalieri method provides unbiased volume estimates of any structure without assumptions about its shape. In this study, we applied the Cavalieri method in combination with point counting to investigate the changes in tumour volume in four patients with high grade glioma, using 3D MRI. In particular, the volume of tumour within the enhancement boundary, the enhancing abnormality (EA), was estimated from T(1) weighted images, and the volume of the non-enhancing abnormality, (NEA) enhancing abnormality, was estimated from T(2) relaxation time and magnetic transfer ratio tissue characterization maps. We compared changes in tumour volume estimated by the Cavalieri method with those obtained using the calliper method. Absolute tumour volume differed significantly between the two methods. Analysis of relative change in tumour volume, based on the WHO criteria, provided a different classification using the calliper and Cavalieri methods. The benefit of the Cavalieri method over the calliper method in the estimation of tumour volume is justified by the following factors. First, Cavalieri volume estimates are mathematically unbiased. Second, the Cavalieri method is highly efficient under an appropriate sampling density (i.e. EA volume estimates can be obtained with a coefficient of error no higher than 5% in 2-3 min). Third, the source of variation of the volume estimates due to disagreements between observers, and within observer, is much greater in the positioning of the calliper diameters than in the identification of the tumour boundaries when applying the Cavalieri method. Additionally, the error prediction formula, available to estimate the coefficient of error of Cavalieri volume estimates from the data, allows us to establish more precise classification criteria against which to identify potentially clinical significant changes in tumour volume.

A computer-aided diagnosis for locating abnormalities in bone scintigraphy by a fuzzy system with a three-step minimization approach

Bone scintigraphy is an effective method to diagnose bone diseases such as bone tumors. In the scintigraphic images, bone abnormalities are widely scattered on the whole body. Conventionally, radiologists visually check the whole-body images and find the distributed abnormalities based on their expertise. This manual process is time-consuming and it is not unusual to miss some abnormalities. In this paper, a computer-aided diagnosis (CAD) system is proposed to assist radiologists in the diagnosis of bone scintigraphy. The system will provide warning marks and abnormal scores on some locations of the images to direct radiologists' attention toward these locations. A fuzzy system called characteristic-point-based fuzzy inference system (CPFIS) is employed to implement the diagnosis system and three minimizations are used to systematically train the CPFIS. Asymmetry and brightness are chosen as the two inputs to the CPFIS according to radiologists' knowledge. The resulting CAD system is of a small-sized rule base such that the resulting fuzzy rules can be not only easily understood by radiologists, but also matched to and compared with their expert knowledge. The prototype CAD system was tested on 82 abnormal images and 27 normal images. We employed free-response receiver operating characteristics method with the mean number of false positives (FPs) and the sensitivity as performance indexes to evaluate the proposed system. The sensitivity is 91.5% (227 of 248) and the mean number of FPs is 37.3 per image. The high sensitivity and moderate numbers of FP marks per image shows that the proposed method can provide an effective second-reader information to radiologists in the diagnosis of bone scintigraphy.

MAP MRF joint segmentation and registration of medical images

The problems of segmentation and registration are traditionally approached individually, yet the accuracy of one is of great importance in influencing the success of the other. In this paper, we aim to show that more accurate and robust results may be obtained through seeking a joint solution to these linked processes. The outlined approach applies Markov random fields in the solution of a maximum a posteriori model of segmentation and registration. The approach is applied to synthetic and real MRI data.

Segmentation of magnetic resonance brain images through discriminant analysis

Segmentation (tissue classification) of medical images obtained from a magnetic resonance (MR) system is a primary step in most applications of medical image post-processing. This paper describes nonparametric discriminant analysis methods to segment multispectral MR images of the brain. Starting from routinely available spin-lattice relaxation time, spin-spin relaxation time, and proton density weighted images (T1w, T2w, PDw), the proposed family of statistical methods is based on: (i) a transform of the images into components that are statistically independent from each other; (ii) a nonparametric estimate of probability density functions of each tissue starting from a training set; (iii) a classic Bayes 0-1 classification rule. Experiments based on a computer built brain phantom (brainweb) and on eight real patient data sets are shown. A comparison with parametric discriminant analysis is also reported. The capability of nonparametric discriminant analysis in improving brain tissue classification of parametric methods is demonstrated. Finally, an assessment of the role of multispectrality in classifying brain tissues is discussed.

Normalization of brain magnetic resonance images using histogram even-order derivative analysis

The even-ordered (2nd, 4th and 6th) derivatives of a brain MRI histogram were used to calculate a characteristic value for white matter, which was used to normalize the image intensity scale. Simulated image histograms were used to estimate the methodological error as a function of noise level, and the optimum derivative order was determined for each image type studied (T1-, T2- and density-weighted). The algorithm yielded highly reproducible results when used in conjunction with a threshold-sensitive brain segmentation algorithm. It also proved insensitive to the presence of extra-cranial tissues. This method of histogram analysis could find utility in a variety of applications that demand robust intensity normalization including image registration, brain segmentation, tissue classification and spatial inhomogeneity correction.

Segmentation and visualization of nasopharyngeal carcinoma using MRI

In this study, a semi-automatic system was developed for nasopharyngeal carcinoma (NPC) tumor segmentation, volume measurement and visualization using magnetic resonance imaging (MRI). Some novel algorithms for tumor segmentation from MRI and inter-slice interpolation were integrated in this medical diagnosis system. This system was applied to 10 MR image data sets of NPC patients and satisfactory results were achieved. This system can be used as a clinical image analysis tool for doctors or radiologists to obtain tumor location from MRI, tumor volume estimation, and 3D information.

Detection of spectral signatures in multispectral MR images for classification

This paper presents a new spectral signature detection approach to magnetic resonance (MR) image classification. It is called constrained energy minimization (CEM) method, which is derived from the minimum variance distortionless response in passive sensor array processing. It considers a bank of spectral channels as an array of sensors where each spectral channel represents a sensor and object spectral signature in multispectral MR images are viewed as signals impinging upon the array. The strength of the CEM lies on its ability in detection of spectral signatures of interest without knowing image background. The detected spectral signatures are then used for classification. The CEM makes use of a finite impulse response (FIR) filter to linearly constrain a desired object while minimizing interfering effects caused by other unknown signal sources. Unlike most spatial-based classification techniques, the proposed CEM takes advantage of spectral characteristics to achieve object detection and classification. A series of experiments is conducted and compared with the commonly used c-means method for performance evaluation. The results show that the CEM method is a promising and effective spectral technique for MR image classification.

Orthogonal subspace projection-based approaches to classification of MR image sequences

Orthogonal subspace projection (OSP) approach has shown success in hyperspectral image classification. Recently, the feasibility of applying OSP to multispectral image classification was also demonstrated via SPOT (Satellite Pour 1'Observation de la Terra) and Landsat (Land Satellite) images. Since an MR (magnetic resonance) image sequence is also acquired by multiple spectral channels (bands), this paper presents a new application of OSP in MR image classification. The idea is to model an MR image pixel in the sequence as a linear mixture of substances (such as white matter, gray matter, cerebral spinal fluid) of interest from which each of these substances can be classified by a specific subspace projection operator followed by a desired matched filter. The experimental results show that OSP provides a promising alternative to existing MR image classification techniques.

Increased differentiation of intracranial white matter lesions by multispectral 3D-tissue segmentation: preliminary results

MRI is a very sensitive imaging modality, however with relatively low specificity. The aim of this work was to determine the potential of image post-processing using 3D-tissue segmentation technique for identification and quantitative characterization of intracranial lesions primarily in the white matter. Forty subjects participated in this study: 28 patients with brain multiple sclerosis (MS), 6 patients with subcortical ischemic vascular dementia (SIVD), and 6 patients with lacunar white matter infarcts (LI). In routine MR imaging these pathologies may be almost indistinguishable. The 3D-tissue segmentation technique used in this study was based on three input MR images (T(1), T(2)-weighted, and proton density). A modified k-Nearest-Neighbor (k-NN) algorithm optimized for maximum computation speed and high quality segmentation was utilized. In MS lesions, two very distinct subsets were classified using this procedure. Based on the results of segmentation one subset probably represent gliosis, and the other edema and demyelination. In SIVD, the segmented images demonstrated homogeneity, which differentiates SIVD from the heterogeneity observed in MS. This homogeneity was in agreement with the general histological findings. The LI changes pathophysiologically from subacute to chronic. The segmented images closely correlated with these changes, showing a central area of necrosis with cyst formation surrounded by an area that appears like reactive gliosis. In the chronic state, the cyst intensity was similar to that of CSF, while in the subacute stage, the peripheral rim was more prominent. Regional brain lesion load were also obtained on one MS patient to demonstrate the potential use of this technique for lesion load measurements. The majority of lesions were identified in the parietal and occipital lobes. The follow-up study showed qualitatively and quantitatively that the calculated MS load increase was associated with brain atrophy represented by an increase in CSF volume as well as decrease in "normal" brain tissue volumes. Importantly, these results were consistent with the patient's clinical evolution of the disease after a six-month period. In conclusion, these results show there is a potential application for a 3D tissue segmentation technique to characterize white matter lesions with similar intensities on T(2)-weighted MR images. The proposed methodology warrants further clinical investigation and evaluation in a large patient population.

Three-dimensional segmentation of anatomical structures in MR images on large data bases

In this paper an image-based method founded on mathematical morphology is presented in order to facilitate the segmentation of cerebral structures over large data bases of 3D magnetic resonance images (MRIs). The segmentation is described as an immersion simulation, applied to the modified gradient image, modeled by a generated 3D-region adjacency graph (RAG). The segmentation relies on two main processes: homotopy modification and contour decision. The first one is achieved by a marker extraction stage where homogeneous 3D-regions are identified. This stage uses contrasted regions from morphological reconstruction and labeled flat regions constrained by the RAG. Then, the decision stage intends to precisely locate the contours of regions detected by the marker extraction. This decision is performed by a 3D extension of the watershed transform. The method has been applied on a data base of 3D brain MRIs composed of fifty patients. Results are illustrated by segmenting the ventricles, corpus callosum, cerebellum, hippocampus, pons, medulla and midbrain on our data base and the approach is validated on two phantom 3D MRIs.

Automatic segmentation of non-enhancing brain tumors in magnetic resonance images

Tumor segmentation from magnetic resonance (MR) images may aid in tumor treatment by tracking the progress of tumor growth and/or shrinkage. In this paper we present the first automatic segmentation method which separates non-enhancing brain tumors from healthy tissues in MR images to aid in the task of tracking tumor size over time. The MR feature images used for the segmentation consist of three weighted images (T1, T2 and proton density (PD)) for each axial slice through the head. An initial segmentation is computed using an unsupervised fuzzy clustering algorithm. Then, integrated domain knowledge and image processing techniques contribute to the final tumor segmentation. They are applied under the control of a knowledge-based system. The system knowledge was acquired by training on two patient volumes (14 images). Testing has shown successful tumor segmentations on four patient volumes (31 images). Our results show that we detected all six non-enhancing brain tumors, located tumor tissue in 35 of the 36 ground truth (radiologist labeled) slices containing tumor and successfully separated tumor regions from physically connected CSF regions in all the nine slices. Quantitative measurements are promising as correspondence ratios between ground truth and segmented tumor regions ranged between 0.368 and 0.871 per volume, with percent match ranging between 0.530 and 0.909 per volume.

Random-grid stereologic volumetry of MR head scans

Point count stereology is a useful tool in obtaining volumetric measures of objects in three-dimensional (3D) images when the segmentation of objects is not feasible. Presently, fixed-grid 3D stereology is being used where a 3D parallelepiped grid is randomly placed for sampling the image space in order to generate test points. Although this is a popular technique, the use of a fixed grid introduces errors in the final estimate in practice and makes the technique inefficient. Random-grid 3D stereology is introduced to improve the efficiency of the volume estimates in stereology. In this manuscript, we prove random-grid stereology as a more consistent technique than fixed-grid stereology and use it for volumetry of the brain and ventricles in magnetic resonance (MR) head scans. We demonstrate superior efficiency and accuracy of random-grid stereology with experiments. Also, the effects of grid sizes, the optimal directions of sectioning the object for volume estimates of the brain and ventricles, and the reliability of the technique are investigated. J. Magn. Reson. Imaging 2000;12:833-841.

Determination of intracranial tumor volumes in a rodent brain using magnetic resonance imaging, Evans blue, and histology: a comparative study

The measurement of tumor volumes is a practical and objective method of assessing the efficacy of a therapeutic agent. However, the relative accuracy of different methods of assessing tumor volume has been unclear. Using T1-weighted, gadolinium-enhanced magnetic resonance Imaging (T1-MRI), Evans Blue infusion and histology we measured intracranial tumor volumes in a rodent brain tumor model (RT2) at days 10, 16 and 18 after implantation of cells in the caudate putamen. There is a good correlation between tumor volumes comparing T1-MRI and Evans Blue (r2 = 0.99), T1-MRI and Histology (r2 = 0.98) and histology and Evans Blue (r2 = 0.93). Each of these methods is reliable in estimating tumor volumes in laboratory animals. There was significant uptake of gadolinium and Evans Blue in the tumor suggesting a wide disruption of the blood-brain barrier.

Virtual endoscopic environments in modern neurosurgical practice

Modern radiographic techniques have allowed the creation of high-definition planar images that can provide important anatomical as well as physiological data. Planar imaging sets can be reformatted into three-dimensional (3-D) data sets that can then be manipulated to demonstrate important anatomical or gross pathological features. Three-dimensional data sets have been used with success in modern image-guided or frameless stereotactic surgery. Another potential application is so-called "virtual endoscopy" or "scopeless endoscopy," in which a 3-D anatomical data set is reformatted into a volume-rendered image that can then be viewed. By reformatting images in this way, a "surgeon's-eye" view can be obtained, which can aid in presurgical planning and diagnosis. The use of virtual endoscopy has the potential to increase our understanding of the appropriate anatomy and the anatomical relationships most apparent during neurosurgical approaches. In so doing, virtual endoscopy may serve as an important means of planning for therapeutic interventions.

On the other hand, one must always be cognizant of the technical limitations of these studies regardless of the quality of the reconstructed images. Prospective, correlative, clinical studies in which the anatomical advantages of virtual-based endoscopy are evaluated in large cadaver or patient series must be performed. Until then, the only potential ways to compensate for errors that exist in the algorithms and reconstructions of 3-D endoscopic images are based on the surgeon's understanding of the clinical state of the patient and prior experience with the anatomy in the region of question.

Feature extraction for MRI segmentation

Magnetic resonance images (MRIs) of the brain are segmented to measure the efficacy of treatment strategies for brain tumors. To date, no reproducible technique for measuring tumor size is available to the clinician, which hampers progress of the search for good treatment protocols. Many segmentation techniques have been proposed, but the representation (features) of the MRI data has received little attention. A genetic algorithm (GA) search was used to discover a feature set from multi-spectral MRI data. Segmentations were performed using the fuzzy c-means (FCM) clustering technique. Seventeen MRI data sets from five patients were evaluated. The GA feature set produces a more accurate segmentation. The GA fitness function that achieves the best results is the Wilks's lambda statistic when applied to FCM clusters. Compared to linear discriminant analysis, which requires class labels, the same or better accuracy is obtained by the features constructed from a GA search without class labels, allowing fully operator independent segmentation. The GA approach therefore provides a better starting point for the measurement of the response of a brain tumor to treatment.

Review and evaluation of MRI nonuniformity corrections for brain tumor response measurements

Current MRI nonuniformity correction techniques are reviewed and investigated. Many approaches are used to remedy this artifact, but it is not clear which method is the most appropriate in a given situation, as the applications have been with different MRI coils and different clinical applications. In this work four widely used nonuniformity correction techniques are investigated in order to assess the effect on tumor response measurements (change in tumor volume over time): a phantom correction method, an image smoothing technique, homomorphic filtering, and surface fitting approach. Six brain tumor cases with baseline and follow-up MRIs after treatment with varying degrees of difficulty of segmentation were analyzed without and with each of the nonuniformity corrections. Different methods give significantly different correction images, indicating that rf nonuniformity correction is not yet well understood. No improvement in tumor segmentation or in tumor growth/shrinkage assessment was achieved using any of the evaluated corrections.

MRI measurement of brain tumor response: comparison of visual metric and automatic segmentation

An automatic magnetic resonance imaging (MRI) multispectral segmentation method and a visual metric are compared for their effectiveness to measure tumor response to therapy. Automatic response measurements are important for multicenter clinical trials. A visual metric such as the product of the largest diameter and the largest perpendicular diameter of the tumor is a standard approach, and is currently used in the Radiation Treatment Oncology Group (RTOG) and the Eastern Cooperative Oncology Group (EGOG) clinical trials. In the standard approach, the tumor response is based on the percentage change in the visual metric and is categorized into cure, partial response, stable disease, or progression. Both visual and automatic methods are applied to six brain tumor cases (gliomas) of varying levels of segmentation difficulty. The analyzed data were serial multispectral MR images, collected using MR contrast enhancement. A fully automatic knowledge guided method (KG) was applied to the MRI multispectral data, while the visual metric was taken from the MRI films using the T1 gadolinium enhanced image, with repeat measurements done by two radiologists and two residents. Tumor measurements from both visual and automatic methods are compared to "ground truth," (GT) i.e., manually segmented tumor. The KG method was found to slightly overestimate tumor volume, but in a consistent manner, and the estimated tumor response compared very well to hand-drawn ground truth with a correlation coefficient of 0.96. In contrast, the visually estimated metric had a large variation between observers, particularly for difficult cases, where the tumor margins are not well delineated. The inter-observer variation for the measurement of the visual metric was only 16%, i.e., observers generally agreed on the lengths of the diameters. However, in 30% of the studied cases no consensus was found for the categorical tumor response measurement, indicating that the categories are very sensitive to variations in the diameter measurements. Moreover, the method failed to correctly identify the response in half of the cases. The data demonstrate that automatic 3D methods are clearly necessary for objective and clinically meaningful assessment of tumor volume in single or multicenter clinical trials.