BJR
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
First published online August 6, 2007
British Journal of Radiology (2007) 80, 609-616
© 2007 British Institute of Radiology
doi: 10.1259/bjr/13992649
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Michail, G
Right arrow Articles by Panayiotakis, G
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Michail, G
Right arrow Articles by Panayiotakis, G

Full paper

Texture analysis of perimenopausal and post-menopausal endometrial tissue in grayscale transvaginal ultrasonography

G Michail, MD 1 A Karahaliou, MSC 2 S Skiadopoulos, PHD 2 C Kalogeropoulou, MD, PHD 3 G Terzis, MD 3 I Boniatis, MSC 2 L Costaridou, PHD 2 G Kourounis, MD, PHD 1 and G Panayiotakis, PHD 2

1 Department of Obstetrics and Gynecology, 2 Department of Medical Physics, 3 Department of Radiology, School of Medicine, University of Patras, 265 00 Patras, Greece

Correspondence: Professor George Panayiotakis, Professor of Medical Physics, Department of Medical Physics, University of Patras, School of Medicine, Patras 265 00, Greece. E-mail: panayiot{at}upatras.gr


   Abstract
Top
 Abstract
Introduction
 Methods and materials
 Results
 Discussion
 References
 
The aim of this study was to investigate the feasibility of texture analysis in characterizing endometrial tissue as depicted in two-dimensional (2D) grayscale transvaginal ultrasonography. Digital transvaginal ultrasound endometrial images were acquired from 65 perimenopausal and post-menopausal women prior to gynaecological operations; histology revealed 15 malignant and 50 benign cases. Images were processed with a wavelet-based contrast enhancement technique. Three regions of interest (ROIs) were identified (endometrium, endometrium plus adjacent myometrium, layer containing endometrial–myometrial interface) on each original and processed image. 32 textural features were extracted from each ROI employing first and second order statistics texture analysis algorithms. Textural features-based models were generated for differentiating benign from malignant endometrial tissue using stepwise logistic regression analysis. Models' performance was evaluated by means of receiver operating characteristic (ROC) analysis. The best logistic regression model comprised seven textural features extracted from the ROIs determined on the processed images; three features were extracted from the endometrium, while four features were extracted from the layer containing the endometrial–myometrial interface. The area under the ROC curve (Az) was 0.956±0.038, providing 86.0% specificity at 93.3% sensitivity using the cut-off level of 0.5 for probability of malignancy. Texture analysis of 2D grayscale transvaginal ultrasound images can effectively differentiate malignant from benign endometrial tissue and may contribute to computer-aided diagnosis of endometrial cancer.


   Introduction
Top
 Abstract
 Introduction
Methods and materials
 Results
 Discussion
 References
 
Endometrial cancer represents one of the most prevalent malignant neoplasias among women [1], with post-menopausal bleeding (PMB) its most common symptom. However, only 10–15% of patients presenting with this symptom actually harbour an endometrial cancer [2], while the incidence of endometrial carcinoma among patients with perimenopausal bleeding may be even less (<2%) [3]. In most patients, clinical decisions such as whether to perform a biopsy or alter hormone replacement regimens are based on sonographic findings [2].

Transvaginal sonography (TVS) represents the cornerstone of the available diagnostic procedures for excluding endometrial carcinoma, mainly due to its non-invasiveness and safe clinical profile. The incidence of suboptimally visualized endometria at TVS has diminished markedly over the past decades as sonographic equipment has evolved, particularly with the development of multifrequency endovaginal transducers and harmonic imaging [3]. While Doppler [4] and sonohysterography techniques [5] enhance endometrial assessment, in conventional grayscale TVS, measurement of double layer endometrial thickness remains the most widely accepted quantitative parameter. Despite the increased performance of TVS in excluding disease, it is considerably less accurate in establishing definite pathology in the presence of thickened endometrium [2].

Several authors have demonstrated the value of endometrial morphology in addition to the measurement of double layer endometrial thickness, particularly in the assessment of the 4–10 mm endometrial thickness "grey zone" [6, 7]. These observations are congruent with the perception that the combined consideration of morphological characteristics and endometrial thickness increases the specificity and negative predictive value of sonography [3].

With the objective of overcoming subjectivity induced by qualitative assessment, the implementation of automated techniques providing objective morphological characteristics, such as computerized texture analysis, would be beneficial, given the value of morphology in endometrial tissue interpretation.

In digital images, texture reflects tonal (intensities of image pixels) and structural (spatial distribution of pixel intensities) properties [8, 9]. Texture analysis refers to algorithms that quantify texture content that may, or may not, be perceived visually. As medical images capture various properties of biological structures, texture analysis of medical images can provide quantitative metrics relevant to the structure, morphology and status of biological tissues [10].

Texture-based classification schemes have been successfully implemented in a variety of ultrasound applications [11–15]. Computerized TVS assessment of endometrial morphology has been applied mainly in assisted reproduction techniques [16–19]; to the authors' knowledge, computerized texture analysis has not been implemented for diagnosing endometrial malignancies in grayscale TVS.

The aim of this study is to investigate the feasibility of computerized texture analysis in characterizing endometrial tissue as depicted in two-dimensional (2D) grayscale TVS images. Furthermore, the effect of a wavelet-based image processing technique in segmentation and subsequent characterization of endometrial tissue is investigated.


   Methods and materials
Top
 Abstract
 Introduction
 Methods and materials
Results
 Discussion
 References
 
Subjects and image acquisition
82 perimenopausal and post-menopausal women with or without vaginal bleeding scheduled for fractionated dilatation and curettage (D&C) or hysterectomy in the Gynaecological Department, University Hospital of Patras, were recruited. Women were considered perimenopausal if they were older than 40 years and had 3–11 months of amenorrhoea, or increased menstrual irregularity for those without amenorrhoea, in the absence of pregnancy. Women were considered as post-menopausal if they were older than 40 years with at least a 1 year absence of menstruation after exclusion of pregnancy. Hormone replacement therapy and tamoxifen medication, any surgical intervention affecting the endometrium in the previous 10 months, former endometrial ablation and the secretory phase of the cycle in pre-menopausal women all represented exclusion criteria. Day-case patients who had undergone a TVS examination elsewhere, and when a TVS could not be repeated in our tertiary department, were also omitted. The protocol of this study has been approved by our hospital ethics board.

Among the 82 recruited patients, 17 were excluded from the study for various reasons, including suboptimal visualization of the endometrium attributed to medioverted or retroverted uteri (n = 12), cancelled surgery (n = 3) and conservative surgery–myectomy (n = 2). Thus, the study population consisted of 65 women, with an average age of 57.2 years (range 41–80 years). 22 patients were perimenopausal (33.8%) and 43 patients (66.2%) were post-menopausal. Uterine perimenopausal or post-menopausal bleeding was the main indication for surgery (n = 49); other indications included uterine fibroids (n = 7), adnexal masses (n = 2), genital prolapse (n = 2), cervical intraepithelial lesions (n = 2) and miscellaneous (n = 3).

All 65 patients underwent a 2D transvaginal scan on a single ATL HDI 3500 ultrasonic imager (Advanced Technology Laboratories, Bothell, WA), with a multifrequency C 5–9 MHz transvaginal transducer, 48 h or less before the gynaecological operation (D&C or hysterectomy) that provided the histological diagnosis. For patients who underwent D&C prior to hysterectomy, only the TVS examination that was performed prior to D&C was considered. One experienced radiologist performed all examinations. For each patient, scan settings (dynamic range, gain, brightness, contrast and focus) were adjusted to obtain a clear depiction of the endometrium. Following an initial assessment of the entire endometrium in multiple planes, endometrial thickness was measured as the maximal double layer thickness in mid-sagittal section, at the thickest area of the endometrium near the fundus and from the outermost border of both sides of the endometrium. If a hypoechoic halo surrounded the endometrium, it was not included in endometrial thickness measurements. At least three measurements of endometrial thickness were performed, and a mean thickness was calculated. To assess possible alterations in endometrial morphology, additional evaluation of the uterus was undertaken. For subsequent analysis, ultrasonic images, in DICOM format, were digitally recorded with 8 bits pixel depth and 768x576 pixels size on a magneto-optical drive.

A collaborative reading by a radiologist and a gynaecologist, who were blinded to the histological diagnoses, was undertaken in order to select the most representative longitudinal image for each patient in the data set of images illustrating a distinctly visible endometrium from the internal cervical os to the fundus, and minimal shadowing artefacts.

Regarding surgical interventions, 24 patients underwent D&C, 40 abdominal hysterectomy and 1 vaginal hysterectomy. Histological diagnoses revealed malignant pathology in 15 women; in the remaining 50 women, benignity was documented. Concerning the malignant cases, 13 patients were diagnosed with endometrial adenocarcinomas, 1 with leiomyosarcoma and 1 with malignant mixed müllerian tumour – carcinosarcoma. No patient was diagnosed with atypical endometrial hyperplasia. Among benign cases, dysfunctional, proliferative or secretory endometria were detected in 8 patients (of whom 4 were perimenopausal), endometrial polyps in 17, simple or complex typical endometrial hyperplasia in 7 and endometrial atrophy in 18 patients. Fibroids were reported in 16 patients, while adenomyosis was diagnosed in 8 patients; 5 patients had additional benign cervical pathology.

The relationship between endometrial thickness and histological diagnoses is presented in Figure 1Go. Of note is the high proportion of patients falling in the 5–10 mm "grey zone" group. As expected, the greater the endometrial thickness, the higher the incidence of endometrial cancer.


Figure 1
View larger version (12K):
[in this window]
[in a new window]

  		Figure 1. Histogram illustrating the distribution of malignant and benign histological diagnoses with respect to endometrial thickness.

 
Image processing with a wavelet-based algorithm
Ultrasound B-scan images often exhibit intensity inhomogeneities as a result of non-uniform beam attenuation within the body [20]. This characteristic not only complicates visual interpretation of images but also impairs segmentation procedure and image analysis, and particularly the computation of quantitative measurements.

In this study, an image processing technique was employed to enhance the contrast of significant characteristics and to facilitate the segmentation procedure. Each image in the data set was processed with a locally adaptive wavelet-based technique [21, 22]. The method is based on local modification of multiscale gradient magnitude values provided by the redundant dyadic wavelet transform. Contrast enhancement is performed by applying a local linear mapping operator on multiscale gradient magnitude values. The first four frequency scales are used to enhance image characteristics of different size. GoFigures 2 and 3Go represent application examples of the locally adaptive wavelet technique on two images from the data set. Figure 2aGo depicts an endometrial adenocarcinoma presenting with increased endometrial thickness, whereas Figure 3aGo depicts derangement of the uterine contour by a leiomyosarcoma. Both corresponding processed images (GoFigures 2b and 3bGo) illustrate a significant amelioration of endometrial perception.


Figure 2
View larger version (51K):
[in this window]
[in a new window]

  		Figure 2. (a) Original image depicting a case of well-differentiated endometrial adenocarcinoma of endometrioid type. (b) The corresponding processed image illustrates more distinctly the disruption of the subendometrial halo.

 

Figure 3
View larger version (57K):
[in this window]
[in a new window]

  		Figure 3. (a) Original image depicting derangement of the uterine contour by a leiomyosarcoma. Dashed arrows indicate the leiomyosarcoma; solid arrows highlight the minimal distortion of the endometrial contour caused. (b) In the corresponding processed image, visualization of the endometrial–leiomyosarcoma interface has been significantly enhanced.

 
An image visualization tool, developed in our department [23, 24], was used for the application of the processing technique and for the segmentation procedure described in the following section. This tool is domain specific to medical imaging and provides global and adaptive wavelet functionality, in addition to conventional visualization operations.

Endometrial tissue segmentation
A radiologist and a gynaecologist, in consensus, retraced manually two regions of interest (ROIs) on each original and processed image in the data set. They were both blinded to the histological diagnoses. The first region of interest (ROI1) corresponds to the endometrium without including the endometrial border, and the second (ROI2) comprises the endometrium along with the adjacent area of the myometrium (endometrium plus the layer containing the endometrial–myometrial interface). Specifically, ROI1 and ROI2 boundaries were delineated on either side of the endometrial–myometrial interface at approximately equal small distances. In order to investigate the textural properties of the layer that contains the endometrial–myometrial interface, a third region of interest (ROI3) was obtained by subtracting the two manually segmented ROIs (ROI2–ROI1).

Figure 4Go depicts the manually segmented ROIs in the original (a) and the corresponding processed image (b). It is evident that the determined ROIs corresponding to the original and processed images are not identical.


Figure 4
View larger version (56K):
[in this window]
[in a new window]

  		Figure 4. The two manually segmented regions of interest(ROIs) corresponding to the endometrium without including the endometrial border (ROI1) and to endometrium along with the adjacent area of the myometrium (ROI2) in the original (a) and the corresponding processed image (b). The dashed line indicates the endometrial–myometrial interface. Histology documented proliferative endometrium.

 
Feature extraction
Prior to feature extraction, both original and processed images were normalized in order to minimize the dependence of the endometrial sonographic appearance on the patients' parameter settings. Specifically, for each image, the mean grey level value µ and the standard deviation {sigma} were calculated. Normalization was performed by decreasing the grey level value of each pixel by the mean value µ and dividing it by the standard deviation {sigma}. The modified grey level values were rescaled in the range of [0, 255] (8 bits pixel depth) [25].

32 textural features were extracted from each of the determined ROIs (ROI1, ROI2 and ROI3) on both original and processed images, employing first and second order statistics texture analysis algorithms [26–28]. The extracted features are capable of quantifying the inherent textural properties of the determined ROIs.

First order statistics textural features
First order statistics provide different statistical properties (first four statistical moments) of the intensity histogram of an image [26]. They depend only on individual pixel values and not on the interaction or co-occurrence of neighbouring pixel values. In this study, four first order textural features were calculated from each ROI: mean, standard deviation, skewness and kurtosis.

Second order statistics textural features
The Grey Level Co-occurrence Matrix (GLCM) is a well-established robust statistical tool for extracting second order texture information from images [27, 28]. The GLCM characterizes the spatial distribution of grey levels in the selected ROI. An element at location (i,j) of the GLCM signifies the joint probability density of the occurrence of grey levels i and j in a specified orientation {theta} and specified distance d from each other. Thus, for different {theta} and d values, different GLCMs result from each selected ROI.

In this study, four GLCMs, corresponding to four different directions ({theta} = 0°, 45°, 90° and 135°) and one distance (d = 1 pixel), were computed for each selected ROI. 14 features were derived from each GLCM. Specifically, the features studied were angular second moment, contrast, correlation, variance, inverse difference moment, sum average, sum variance, sum entropy, entropy, difference entropy, information measure of correlation 1, information measure of correlation 2, shade and prominence. Four values were obtained for each feature corresponding to the four matrices. The mean and range of these 4 values were calculated, comprising a total of 28 second order textural features.

The 32 (first and second order statistics) textural features extracted from each ROI of the original and processed images were used as input for logistic regression analysis to determine the most discriminating subset of features for differentiation of benign from malignant endometrial tissue. The histological findings of benign or malignant diagnoses were considered as the truth table.

Segmentation reproducibility
In order to assess intraobserver reproducibility in the segmentation procedure, the 130 ROIs (65 ROI1 and 65 ROI2) were blindly delineated twice (after a 2 month period), collaboratively by a radiologist and a gynaecologist, on each original and processed image in the data set. Both radiologist and gynaecologist were blinded to the histological diagnoses. The area overlap criterion between the two segmented ROIs was used, defined as the ratio of intersection to the union of the two segmented areas [29]. Reproducibility was defined by the percentage of cases with an area overlap value exceeding 80%. To assess whether the differences between the two segmented areas were statistically significant, a non-parametric statistical test (two-tailed Wilcoxon signed ranks test [30]) with a significance level of 0.05 was applied.

Tissue characterization
Logistic regression analysis [31] was applied to determine the optimal subset of textural features (descriptors) capable of differentiating benign from malignant endometrial tissue. Specifically, the forward stepwise feature selection method of the logistic regression analysis was performed using a statistical software package (NCSS Statistical Software 2004, Kaysville, UT).

In order to assess the effect of the segmentation procedure (contribution of the various ROIs) in the endometrial tissue characterization task, separate logistic regression models were generated for the original images. Furthermore, aiming to investigate the effect of the processing technique in the characterization task, additional regression models were generated for the processed images.

For each data set (original and processed images), separate models were constructed for each ROI. Specifically, three regression models were constructed, with each one using as inputs the same 32 textural features extracted from ROI1, ROI2 and ROI3, respectively. An additional regression model was constructed combining the 32 textural features extracted from ROI1 and the 32 textural features extracted from ROI3. Hereafter, this model will be referred to as the one corresponding to (ROI1/ROI3).

The aforementioned models were constructed in the form of logit({pi}) equations (logarithm of the odds) as follows:


Formula 001

where Xs are the independent variables (features), v is the number of features (descriptors) in the model, Y is the binary dependent variable (pathology), which has two possible values, y1 (1: malignant) and y2 (0: benign), beta0 is the intercept and betas are the logistic regression coefficients. From these logit({pi}) values, the estimated probability of malignancy ({pi}) for tissue depicted in a ROI can be obtained from the inverse logistic transformation:


Formula 002

The performance of each logistic regression model was evaluated by means of ROC analysis using the area under the receiver operating characteristic (ROC) curve (Az) and the corresponding standard error (SE) [32]. Comparative evaluation was performed among models constructed of features extracted from: (a) original ROIs, (b) processed ROIs and (c) original and the corresponding processed ROIs. Furthermore, the sensitivity level of 93.3% (i.e. which correctly characterized 14 out of 15 malignant endometria) was selected to compare the specificity for each model.


   Results
Top
 Abstract
 Introduction
 Methods and materials
 Results
Discussion
 References
 
Segmentation reproducibility
Reproducibility values in the segmentation of ROI1 and ROI2 on original images were 83% (54/65) and 85% (55/65), respectively; the corresponding reproducibility values on processed images were 92% (60/65) and 91% (59/65). The differences between the two areas in the segmentation of ROI1 and ROI2 were not statistically significant for either original (p = 0.2438 and 0.3054, respectively) or processed images (p = 0.5464 and 0.3396, respectively).

Tissue characterization
Table 1Go provides the best logistic regression models, as chosen by forward stepwise selection of textural features extracted from ROI1, ROI2 and ROI3, as well as for the model combining the features from ROI1 and ROI3 (ROI1/ROI3), in terms of logit({pi}) equations for both original and processed images. All features entered in the regression models are statistically significant (Wald test for each regression coefficient, p<0.05). As observed in Table 1Go, the features included in the regression models are differentiated with respect to the selected ROI, suggesting that each ROI is characterized by different inherent textural properties. Thus, different optimal subsets of textural features (descriptors) can be used to differentiate between malignant and benign tissue for both original and processed images. Furthermore, for both original and processed data sets, the regression model corresponding to the endometrium plus the layer containing the endometrial–myometrial interface (ROI2) is different from the regression model produced by combining features from the same two regions (ROI1/ROI3), suggesting that these two regions possess independent inherent textural properties. Additionally, the textural features included in regression models corresponding to each ROI are differentiated with respect to the original and processed images.


View this table:
[in this window]
[in a new window]

  		Table 1. Best logistic regression models, as chosen by forward stepwise selection of textural features extracted from region of interest(ROI)1, ROI2, ROI3 and ROI1/ROI3, in terms of logit({pi}) equations for both original and processed images

 
Table 2Go provides the Az values and corresponding SEs achieved by the best logistic regression models, as chosen by the forward stepwise selection method, corresponding to ROI1, ROI2, ROI3 and ROI1/ROI3, for both original and processed images. The corresponding ROC curves are presented in Figure 5Go. For the original images data set, the performance of logistic regression models is differentiated with respect to the selected ROI (Figure 5aGo). The best performance is achieved by features extracted from the layer that contains the endometrial–myometrial interface (ROI3). The Az value is 0.901±0.055, providing 68.0% specificity at 93.3% sensitivity using the cut-off level of 0.35 for probability of malignancy. For the processed images data set, the performance of the regression models also depends on the selected ROI (Figure 5bGo). Specifically, the best performance is obtained by the model produced by combining features from the endometrium and the layer encompassing the endometrial–myometrial interface (ROI1/ROI3). The Az value is 0.956±0.038, providing 86.0% specificity at 93.3% sensitivity using the cut-off level of 0.5 for probability of malignancy. Furthermore, the performance of regression models corresponding to each ROI is differentiated with respect to the original and processed images. In particular, the models corresponding to the processed data set demonstrate increased performance with respect to the models from the original data set. The highest improvement in regression models' performance between processed and original ROI is achieved for ROI1, yielding improvement in the Az value of 0.078 and in specificity of 22% (i.e. correctly characterized 11 out of 50 benign endometria) at the 93.3% sensitivity level.


View this table:
[in this window]
[in a new window]

  		Table 2. Area under the receiver operating characteristic(ROC) curve (Az) and standard error (SE) achieved by logistic regression models corresponding to region of interest (ROI)1, ROI2, ROI3 and ROI1/ROI3, for both original and processed images

 

Figure 5
View larger version (21K):
[in this window]
[in a new window]

  		Figure 5. Receiver operating characteristic(ROC) curves corresponding to the logistic regression models extracted from region of interest (ROI)1, ROI2, ROI3 and ROI1/ROI3 for original (a) and processed (b) images.

 

   Discussion
Top
 Abstract
 Introduction
 Methods and materials
 Results
 Discussion
References
 
This paper demonstrates the feasibility of computerized texture analysis in characterizing endometrial tissue, as depicted in 2D grayscale TVS. Each of the determined ROIs possesses independent textural information that could have a bearing on the optimal subset of features capable of differentiating malignant from benign endometrial tissue. Regarding the original images, ROI3 demonstrated the best discriminating performance (Figure 5aGo). This suggests that this layer, besides encompassing information on endometrial–myometrial interface regularity, a possible indicator of myometrial invasion [33], also captures texture of high discriminating power.

The performance of the regression models with respect to tissue characterization was significantly improved when the wavelet-based processing technique was applied prior to the segmentation task. Specifically, the regression model produced by combining features extracted from ROI1/ROI3 of the processed images achieved the highest performance (Figure 5bGo). This high performance may be attributed to: (a) more accurate delineation of ROIs obtained by endometrial contrast enhancement in the processed images; (b) the additionally revealed textural information provided by the processing technique yielding features with higher discriminating ability; (c) the hypothesis of two separate regions, corresponding to ROI1 and ROI3, capturing significant differences in textural information, resulting in improved endometrial characterization compared with merging the two regions into one (reflecting ROI2). As suggested by the proposed method, the endometrium and the endometrial–myometrial interface layer should be analysed separately as distinct regions in processed images. Combining textural features from the distinct regions provides improved performance in endometrial characterization.

It would be useful to compare the results obtained from this study with results reported aiming to increase TVS specificity. However, this might not be feasible because of inhomogeneities in case samples (size, patients' characteristics, etc.) used, the sonographic criteria assessed and variability in the statistical analyses performed [6, 7]. Moreover, there is no published study of computerized characterization of endometrial tissue, as depicted in TVS.

The proposed computerized approach for objective characterization of endometrial tissue has shown promising preliminary results. However, we are aware that our study design suffers from a number of shortcomings: both perimenopausal and post-menopausal women were recruited, and not all presented with vaginal bleeding. This inhomogeneity might reflect non-consistency in endometrial texture, yet the histological reports documented functional endometrium only in a few patients. Furthermore, surgical interventions were not uniform; D&C is undoubtedly inferior to hysterectomy in terms of accuracy of the histological diagnoses [34]. However, we have pooled heterogeneous data in accordance with other related studies [6, 35], due to the small size of case sample.

Additional work is required to improve the overall performance of the proposed method. Efforts will focus on assessing the performance of the proposed method in newer generation ultrasound scanners and larger data sets. The processing technique used in this work was found to be fundamental to the task of tissue characterization. To improve further the image processing technique, a speckle reduction [36] method should be considered prior to contrast enhancement. Finally, we aim to investigate the potential increase in the diagnostic accuracy of TVS by combining computerized textural features with shape features for capturing irregularity variability in the endometrial–myometrial interface.


   Acknowledgments
 
The authors would like to thank the staff of the Department of Radiology at the University Hospital of Patras for their contribution to this work.

Received for publication July 27, 2006. Revision received November 15, 2005. Accepted for publication November 23, 2006.


   References
Top
 Abstract
 Introduction
 Methods and materials
 Results
 Discussion
 References
 

  1. Parker SL, Tong T, Bolden S, Wingo PA. Cancer statistics, 1996. CA Cancer J Clin 1996;46:5–27.[Abstract]
  2. Fleischer AC. New insights into transvaginal sonography of endometrial disorders. Ultrasound Quarterly 2001;17:11–19.[CrossRef][Medline]
  3. Dubinsky TJ. Value of sonography in the diagnosis of abnormal vaginal bleeding. J Clin Ultrasound 2004;32:348–53.[CrossRef][Medline]
  4. Alcazar Jl, Castillo G, Minguez JA, Galan MJ. Endometrial blood flow mapping using transvaginal power Doppler sonography in women with postmenopausal bleeding and thickened endometrium. Ultrasound Obstet Gynecol 2003;21:538–88.
  5. Baldwin MT, Dudiak KM, Marks CA. Focal intracavitary masses recognized with the hyperechoic line sign at endovaginal US and characterized with hysterosonography. Radiographics 1999;19:927–35.[Abstract/Free Full Text]
  6. Weigel M, Friese K, Strittmatter F, Melchert F. Measuring the thickness – is that all we have to do for sonographic assessment of endometrium in postmenopausal women? Ultrasound Obstet Gynecol 1995;6:97–102.[CrossRef][Medline]
  7. Randelzhofer B, Prompeler HJ, Sauerbrei W, Madjar H, Emons G. Value of sonomorphologic criteria of the endometrium in women with postmenopausal bleeding. Ultrasound Obstet Gynecol 2002;19:62–8.[CrossRef][Medline]
  8. Amadasun M, King R. Textural features corresponding to textural properties. IEEE Trans System Man Cybernetics 1998;19:1264–74.
  9. Castellano G, Bonilha L, Li LM, Cendes F. Texture analysis of medical images. Clin Radiol 2004;59:1061–9.[CrossRef][Medline]
  10. Bocchi L, Coppini G, De Dominicis R, Valli G. Tissue characterization from X-ray images. Med Eng Phys 1997;19:336–42.[CrossRef][Medline]
  11. Bader W, Bohmer S, van Leeuwen P, Hackmann J, Westhof G, Hatzmann W. Does texture analysis improve breast ultrasound precision? Ultrasound Obstet Gynecol 2000;15:311–16.[CrossRef][Medline]
  12. Smutek D, Sara R, Sucharda P, Tjahjadi T, Svec M. Image texture analysis of sonograms in chronic inflammations of thyroid gland. Ultrasound Med Biol 2003;29:1531–43.[CrossRef][Medline]
  13. Christodoulou CI, Pattichis CS, Kyriacou E, Pattichis MS, Pantziaris M, Nicolaides A. Texture and morphological analysis of ultrasound images of the carotid plaque for the assessment of stroke. In: Costaridou L, editor. Medical image analysis methods. Boca Raton, FL: Taylor & Francis Group LLC, CRC Press, 2005: 87–135
  14. Yoshida H, Casalino DD, Keserci B, Coskun A, Ozturk O, Savranlar A. Wavelet-packet-based texture analysis for differentiation between benign and malignant liver tumours in ultrasound images. Phys Med Biol 2003;48:3735–53.[CrossRef][Medline]
  15. Mohamed SS, Salama MM, Kamel M, El-Saadany EF, Rizkalla K, Chin J. Prostate cancer multi-feature analysis using trans-rectal ultrasound images. Phys Med Biol 2005;50:175–85.[CrossRef]
  16. Leibovitz Z, Degani S, Rabia R, Tal J, Eibchitz I, Shapiro I, et al. Endometrium-to-myometrium relative echogenicity coefficient. A new sonographic approach for the quantitative assessment of endometrial echogenicity. Gynecol Obstet Invest 1998;45:121–5.[CrossRef][Medline]
  17. Fanchin R, Righini C, Olivennes F, Taieb J, de Ziegler D, Frydman R. Computerized assessment of endometrial echogenicity: clues to the endometrial effects of premature progesterone elevation. Fertil Steril 1999;71:174–81.[CrossRef][Medline]
  18. Fanchin R, Righini C, Ayoubi JM, Olivennes F, de Ziegler D, Frydman R. New look at the endometrial echogenicity: objective computer-assisted measurements predict endometrial receptivity in vitro fertilization–embryo transfer. Fertil Steril 2000;74:274–80.[CrossRef][Medline]
  19. Pierson RA, Kadash ML, Baerwald AR. Computer assisted three dimensional analysis of endometrial ultrasound images across the menstrual cycle. Abstract of the 15th World Congress on Ultrasound in Obstetrics and Gynecology. Ultrasound Obstet Gynecol 2005;26(OC4:05)314–5.
  20. Xiao G, Brady M, Noble JA, Zhang Y. Segmentation of ultrasound B-mode images with intensity inhomogeneity correction. IEEE Trans Med Imaging 2002;21:48–57.[CrossRef][Medline]
  21. Sakellaropoulos P, Costaridou L, Panayiotakis G. A wavelet-based spatially adaptive method for mammographic contrast enhancement. Phys Med Biol 2003;48:787–803.[CrossRef][Medline]
  22. Costaridou L, Sakellaropoulos P, Skiadopoulos S, Panayiotakis G. Locally adaptive wavelet contrast enhancement. In: Costaridou L, editor. Medical image analysis methods. Boca Raton, FL: Taylor & Francis Group LLC, CRC Press, 2005: 225–57
  23. Sakellaropoulos P, Costaridou L, Panayiotakis G. An image visualization tool in mammography. Med Inform 1999;24:53–73.
  24. Sakellaropoulos P, Costaridou L, Panayiotakis G. Using component technologies for web based wavelet enhanced mammographic image visualization. Med Inform 2000;25:171–81.
  25. Baker S, Nayar SK, Murase H. Parametric feature detection. Int J Comp Vision 1998;27:27–50.[CrossRef]
  26. Gonzalez RC, Woods RE, editors. Representation and description. In: Digital image processing, 2nd edition. Upper Saddle River, NJ: Prentice-Hall, 2002: 643–92
  27. Haralick RM, Shanmugam K, Dinstein I. Textural features for image classification. IEEE Trans System Man Cybernetics 1973;SMC-3:610–21.
  28. Walker RF, Jackway P, Longstaff ID. Improving co-occurrence matrix feature discrimination. In: Proceedings of the 3rd Conference on Digital Image Computing: Techniques and Applications, 6–8 December 1995;643–8
  29. te Brake GM, Karssemeijer N. Segmentation of suspicious densities in digital mammograms. Med Phys 2001;28:259–66.[CrossRef][Medline]
  30. Siegel S, Castellan NJ, editors. The case of one sample, two measures or paired replicates. In: Nonparametric statistics for the behavioral sciences, 2nd edition. New York: McGraw-Hill Book Co., 1988: 73–101
  31. Menard S. Applied logistic regression analysis. In: Quantitative applications in the social sciences, No. 106, 2nd edition. Thousand Oaks, CA: Sage Publications, 2002
  32. van Erkel AR, Pattynama PM. Receiver operating characteristic (ROC) analysis: basic principles and applications in radiology. Eur J Radiol 1998;27:88–94.[CrossRef][Medline]
  33. Cagnazzo G, D'Addario V, Martinelli G, Lastilla G. Depth of myometrial invasion in endometrial cancer: preoperative assessment by transvaginal ultrasonography and magnetic resonance imaging. Ultrasound Obstet Gynecol 1992;2:40–3.[CrossRef][Medline]
  34. Stovall TG, Solomon SK, Ling FW. Endometrial sampling prior to hysterectomy. Obstet Gynecol 1989;73:405–9.[Medline]
  35. Dubinsky TJ, Stroehlein K, Abu-Ghazzeh Y, Parvey HR, Maklad N. Prediction of benign and malignant endometrial disease: hysterosonographic–pathologic correlation. Radiology 1999;210:393–7.[Abstract/Free Full Text]
  36. Gupta S, Chauhan RC, Sexana SC. Wavelet-based statistical approach for speckle reduction in medical ultrasound images. Med Biol Eng Comput 2004;42:189–92.[CrossRef][Medline]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Michail, G
Right arrow Articles by Panayiotakis, G
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Michail, G
Right arrow Articles by Panayiotakis, G

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
BJR DMFR IMAGING  ALL BIR JOURNALS