PubMed Health⌕ Search

Biomedical subjects

N Karssemeijer

Publications and source records attributed to N Karssemeijer.

At least 19 recordsLinked to original sources
PubMed2006

Use of border information in the classification of mammographic masses.

We are developing a new method to characterize the margin of a mammographic mass lesion to improve the classification of benign and malignant masses. Towards this goal, we designed features that measure the degree of sharpness and microlobulation of mass margins. We calculated these features in a border region of the mass defined as a thin band along the mass contour. The importance of these features in the classification of benign and malignant masses was studied in relation to existing features used for mammographic mass detection. Features were divided into three groups, each representing a different mass segment: the interior region of a mass, the border and the outer area. The interior and the outer area of a mass were characterized using contrast and spiculation measures. Classification was done in two steps. First, features representing each of the three mass segments were merged into a neural network classifier resulting in a single regional classification score for each segment. Secondly, a classifier combined the three single scores into a final output to discriminate between benign and malignant lesions. We compared the classification performance of each regional classifier and the combined classifier on a data set of 1076 biopsy proved masses (590 malignant and 486 benign) from 481 women included in the Digital Database for Screening Mammography. Receiver operating characteristic (ROC) analysis was used to evaluate the accuracy of the classifiers. The area under the ROC curve (A(z)) was 0.69 for the interior mass segment, 0.76 for the border segment and 0.75 for the outer mass segment. The performance of the combined classifier was 0.81 for image-based and 0.83 for case-based evaluation. These results show that the combination of information from different mass segments is an effective approach for computer-aided characterization of mammographic masses. An advantage of this approach is that it allows the assessment of the contribution of regions rather than individual features. Results suggest that the border and the outer areas contained the most valuable information for discrimination between benign and malignant masses.

Algorithms↗
PubMed2006

Computer aided detection of masses in mammograms as decision support.

Performance of a computer aided detection (CAD) system for masses in mammograms was investigated. Using data collected in an observer study, in which experienced screening radiologists read a series of 500 screening mammograms without CAD, performance of radiologists was compared to the standalone performance of the CAD system. Due to a larger number of FPs (false positives), the performance of CAD was lower than that of the readers. However, when analysis was restricted to mammographic regions identified by the radiologists, it was found that the CAD system was comparable to the readers in discriminating these regions in cancer and non-cancer. In a retrospective analysis, the effect of independent combination of reader scores with CAD was compared to independent combination of scores of two radiologists. No significant difference was found between the results of these two methods. Both methods improved single reading results significantly.

Algorithms↗
PubMed2004

Accuracy of rigid CT-FDG-PET image registration of the liver.

Diagnostic and surgical strategies could benefit from accurate localization of liver malignancies via CT-FDG-PET image registration. However, registration uncertainty occurs due to protocol differences in data-acquisition, the limited spatial resolution of positron emission tomography (PET) and the low uptake of 18F-fluorodeoxyglucose (FDG) in normal liver tissue. To assess this uncertainty, methods were presented to estimate registration precision and systematic bias. A semi-automatic, organ-focused method was investigated to minimize the uncertainty well beyond the typical uncertainty of 5-10 mm obtained by commonly available methods. By restricting registration to the liver region and by isolating the liver on computed tomography (CT) from surrounding structures using a thresholding technique, registration was achieved using the mutual information-based method as implemented in insight toolkit (ITK). CT and FDG-PET images of 10 patients with liver metastases were registered rigidly a number of times. Results of the organ-focused method were compared to results of three commonly available methods (a manual, a landmark-based and a 'standard' mutual information-based method) where no dedicated image processing was performed. The proposed method outperformed the other methods with a precision (mean+/-s.d.) of 2.5+/-1.3 mm and a bias of 1.9 mm with a 95% CI of [1.0, 2.8] mm. Unlike the commonly available methods, our approach allows for robust CT-FDG-PET registration of the liver, with an accuracy better than the spatial resolution of the PET scanner that was used.

Fluorodeoxyglucose F18↗
PubMed2001

Segmentation of suspicious densities in digital mammograms.

State-of-the-art algorithms for detection of masses in mammograms are very sensitive but they also detect many normal regions with slightly suspicious features. Based on segmentations of detected regions, shape and intensity features can be computed that discriminate between normal and abnormal regions. These features can be used to discard false positive detections and hence improve the specificity of the detection method. In this work two different methods to segment suspect regions were examined. A number of different implementations of a region growing method were compared to a discrete dynamic contour method. Both methods were applied to a consecutive data set of 132 mammograms containing masses and architectural distortions, taken from the Dutch screening program. Evaluation of the performance of the methods was done in two different ways. In the first experiment, the segmentations of masses were compared to annotations made by the radiologist. In the second experiment, a number of features were computed for all segmented areas, normal and abnormal, based on which regions were classified with a neural network. The most sophisticated region growing method and the method using the dynamic contour model had a similar performance when evaluation was based on the overlap of the annotations. The second experiment showed that the contours generated by the discrete dynamic contour model were more suited for computation of discriminating features. Contrast features were especially useful to improve the performance of the detection method.

Algorithms↗
PubMed2000

An automatic method to discriminate malignant masses from normal tissue in digital mammograms.

Specificity levels of automatic mass detection methods in mammography are generally rather low, because suspicious looking normal tissue is often hard to discriminate from real malignant masses. In this work a number of features were defined that are related to image characteristics that radiologists use to discriminate real lesions from normal tissue. An artificial neural network was used to map the computed features to a measure of suspiciousness for each region that was found suspicious by a mass detection method. Two data sets were used to test the method. The first set of 72 malignant cases (132 films) was a consecutive series taken from the Nijmegen screening programme, 208 normal films were added to improve the estimation of the specificity of the method. The second set was part of the new DDSM data set from the University of South Florida. A total of 193 cases (772 films) with 372 annotated malignancies was used. The measure of suspiciousness that was computed using the image characteristics was successful in discriminating tumours from false positive detections. Approximately 75% of all cancers were detected in at least one view at a specificity level of 0.1 false positive per image.

Breast Neoplasms↗
PubMed2000

Normalization of local contrast in mammograms.

Equalizing image noise has been shown to be an important step in automatic detection of microcalcifications in digital mammograms. In this study, an accurate adaptive approach for noise equalization is presented and investigated. No additional information obtained from phantom recordings is involved in the method, which makes the approach robust and independent of film type and film development characteristics. Furthermore, it is possible to apply the method on direct digital mammograms as well. In this study, the adaptive approach is optimized by investigating a number of alternative approaches to estimate the image noise. The estimation of high-frequency noise as a function of the grayscale is improved by a new technique for dividing the grayscale in sample intervals and by using a model for additive high-frequency noise. It is shown that the adaptive noise equalization gives substantially better detection results than does a fixed noise equalization. A large database of 245 digitized mammograms with 341 clusters was used for evaluation of the method.

Artifacts↗
PubMed2000

Automated classification of clustered microcalcifications into malignant and benign types.

The objectives in this study were to design and test a fully automated method for classification of microcalcification clusters into malignant and benign types, and to compare the method's performance with that of radiologists. A novel aspect of the approach is that the relative location and orientation of clusters inside the breast was taken into account for feature calculation. Furthermore, correspondence of location of clusters in mediolateral oblique (MLO) and cranio-caudal (CC) views, was used in feature calculation and in final classification. Initially, microcalcifications were automatically detected by using a statistical method based on Bayesian techniques and a Markov random field model. To determine malignancy or benignancy of a cluster, a method based on two classification steps was developed. In the first step, classification of clusters was performed and in the second step a patient based classification was done. A total of 16 features was used in the study. To identify meaningful features, a feature selection was applied, using the area under the receiver operating characteristic (ROC) curve (Az value) as a criterion. For classification the k-nearest-neighbor method was used in a leave-one-patient-out procedure. A database of 192 mammograms with 280 true positive detected microcalcification clusters was used for evaluation of the method. The set consisted of cases that were selected for diagnostic work up during a 4 year period of screening in the Nijmegen region (The Netherlands). Because of the high positive predictive value in the screening program (50%), this set did not contain obvious benign cases. The method's best patient-based performance on this set corresponded with Az = 0.83, using nine features. A subset of the data set, containing mammograms from 90 patients, was used for comparing the computer results to radiologists' performance. Ten radiologists read these cases on a light-box and assessed the probability of malignancy for each patient. All participants had experience in clinical mammography and participated in our observer study during the last 2 days of a 2-week training session leading to screening mammography certification. Results on the subset showed that the method's performance (Az = 0.83) was considerably higher than that of the radiologists (Az = 0.63).

Breast Neoplasms↗
PubMed1999

[Reading screening mammograms with the help of neural networks].

With digital mammography it is possible to assist radiologists in breast cancer screening with computers to improve their reading performance. The need for this has been demonstrated by studies showing a large variability in skill of radiologists reading mammograms. Moreover, retrospective studies show that a significant number of cancers are clearly visible on earlier screening mammograms, even for 'trained' computers. Methods for automated detection of breast cancer in mammograms often use artificial neural networks. These are 'trained' to recognize abnormal mammographic areas using a large database of known cases. For detection of microcalcification clusters very reliable algorithms exist, with such high sensitivity that radiologists can limit their search to areas that have been marked 'suspect' by the computer. The development of methods to recognize malignant masses is much more difficult, but ample progress has been achieved in recent years.

Adult↗
PubMed1999

Changes in mammographic breast density and concomitant changes in breast cancer risk.

Among participants of the biennial Nijmegen breast cancer screening programme, we examined whether diminution of mammographic breast density lowered breast cancer risk. Post-menopausal breast cancer cases (n = 108), who had to have participated in all the five screening rounds prior to their diagnosis, were matched to 400 controls on year of birth and screening history. Controls had to be free of breast cancer at the time of the case's diagnosis. Changes in breast density were measured over a 10-year period, by a fully computerized method. Women in whom 5-25% or >25% of the breast was composed of fibro-glandular density showed a threefold increased 10-year risk compared to women with <5% density. In women with 5-25% density initially, we observed a trend of decreasing risk with diminishing density: when women with <5% density throughout the whole period formed the reference category, the odds ratio (OR) for those who decreased from 5-25% to <5% density was 1.9 [95% confidence interval (CI) = 0.6-6.1] in contrast to the OR of 5.7 (95% CI = 2.2-15.2) for those with persisting 5-25% density. In women who increased from 5-25% density to >25% density the OR was 6.9 (95% CI = 2.1-22.9). In women with >25% density initially, diminishing density was not clearly associated with lowering risk, which may be partly explained by the low number of women who decreased to <5% (n = 12). Due to the limited size of the study these results have to be interpreted with caution. Although the results are not conclusive, they could indicate a trend of decreasing risk with diminishing breast density. Should this effect be real, it may have great implications for the primary prevention of breast cancer or for the identification of high-risk groups who would benefit by more frequent screening. Therefore, large-scale, long-term follow-up studies on the effects of changes in breast density are needed.

Adult↗
PubMed1999

Single and multiscale detection of masses in digital mammograms.

Scale is an important issue in the automated detection of masses in mammograms, due to the range of possible sizes masses can have. In this work, it was examined if detection of masses can be done at a single scale, or whether it is more appropriate to use the output of the detection method at different scales in a multiscale scheme. Three different pixel-based mass-detection methods were used for this purpose. The first method is based on convolution of a mammogram with the Laplacian of a Gaussian, the second method is based on correlation with a model of a mass, and the third is a new approach, based on statistical analysis of gradient-orientation maps. Experiments with simulated masses indicated that little can be gained by applying the methods at a number of scales. These results were confirmed by experiments on a set of 71 cases (132 mammograms) containing a malignant tumor. The performance of each method in a multiscale scheme was similar to the performance at the optimal single scale. A slight improvement was found for the correlation method when the output of different scales was combined. This was especially evident at low specificity levels. The correlation method and the gradient-orientation-analysis method have similar performances. A sensitivity of approximately 75% is reached at a level of one false positive per image. The method based on convolution with the Laplacian of the Gaussian performed considerably worse, in both a single and multiscale scheme.

Aged↗
PubMed1998

Automated classification of parenchymal patterns in mammograms.

A method for automated determination of parenchymal patterns in mammograms has been developed that is insensitive to changes in the mammographic imaging technique. The method was designed to study the relation between breast cancer risk and changes of mammographic density. It includes a new method for automatic segmentation of the pectoral muscle in oblique mammograms, based on application of the Hough transform. The technique developed for classification of parenchymal patterns is based on a distance transform that subdivides the breast tissue area into regions in which distance to the skin line is approximately equal. Features are calculated from grey level histograms computed in these regions. In this way, dependency on varying tissue thickness in the peripheral zone of the breast is minimized. Additional features represent differences between tissue projected in pectoral and breast area. Robustness and classification performance were studied on a test set of 615 digitized mammograms, applying a kNN classifier and leave-one-out for training. Using four density categories in 67% of the cases an exact agreement was obtained with a subjective classification made by a radiologist. The number of cases for which classifications of the radiologist and the program differed by more that one category was only 2%. For more recent mammograms, recorded after 1991, an exact agreement of 80% was obtained.

Automation↗
PubMed1998

Accurate segmentation and contrast measurement of microcalcifications in mammograms: a phantom study.

The authors are developing a computer-aided diagnostic method to assist radiologists in differentiating between malignant and benign clustered microcalcifications in mammograms. In earlier studies we investigated shape and contrast features of microcalcifications for classification. It was found that segmentation strongly influences classification results. For this reason a phantom study has been carried out. The CDMAM phantom, consisting of a pattern of dots with known size and object contrast is used for evaluation of contrast measurement and segmentation. Dots in the range of 0.2-0.8 mm are taken as a model for microcalcifications. In this article performances of different methods for segmentation of microcalcifications are compared. An iterative method based on a Markov random field and a signal dependent criterion give satisfying results. The segmentation performances of both methods are comparable. Also the influence of the modulation transfer function on contrast estimates is determined and effect of exposure level on segmentation is analyzed.

Breast Diseases↗
PubMed1998

Automated detection of breast carcinomas not detected in a screening program.

PURPOSE: To investigate the possibility of automated detection of early signs of cancer that were not detected in a breast cancer screening program. MATERIALS AND METHODS: A set of 75 mammograms (in 65 women) with subtle circumscribed masses, stellate lesions, and architectural distortions that were not detected in a screening program by two radiologists was assembled and extended with 142 normal mammograms (contralateral mammograms in the same 65 women). An automated system for the detection of circumscribed masses and stellate lesions was applied to this set. RESULTS: In 22 (34%) of 65 cases, an early sign of cancer was detected at a specificity of one false-positive finding per image. At a specificity of three false-positive findings per image, 39 (60%) of the cancers were detected. Of the tumors that were classified as screening errors, seven (50%) were found at a specificity of 0.5 false-positive finding per image. CONCLUSION: A substantial proportion of cancers that were missed in a screening program, despite double reading, were found with this detection method at less than one false-positive finding per image.

Algorithms↗
PubMed1997

Computer-assisted reading of mammograms.

Techniques developed in computer vision and automated pattern recognition can be applied to assist radiologists in reading mammograms. With the introduction of direct digital mammography this will become a feasible approach. A radiologist in breast cancer screening can use findings of the computer as a second opinion, or as a pointer to suspicious regions. This may increase the sensitivity and specificity of screening programs, and it may avoid the need for double reading. In this paper methods which have been developed for automated detection of mammographic abnormalities are reviewed. Programs for detecting microcalcification clusters and stellate lesions have reached a level of performance which makes application in practice viable. Current programs for recognition of masses and asymmetry perform less well. Large-scale studies still have to demonstrate if radiologists in a screening situation can deal with the relatively large number of false positives which are marked by computer programs, where the number of normal cases is much higher than in observer experiments conducted thus far.

Breast Neoplasms↗
PubMed1996

Staging urinary bladder cancer after transurethral biopsy: value of fast dynamic contrast-enhanced MR imaging.

PURPOSE: To evaluate contrast enhancement patterns of urinary bladder cancer and surrounding structures and to evaluate a fast dynamic first-pass magnetic resonance (MR) imaging technique in tumor and node staging and in differentiation of urinary bladder cancer from postbiopsy effects. MATERIALS AND METHODS: Sixty-one consecutive patients with histologically proved urinary bladder cancer were referred to undergo unenhanced and dynamic MR imaging 1-4 weeks after transurethral resection or biopsy. Subtraction and time (to beginning of enhancement) images were acquired. RESULTS: Results with unenhanced T1- and T2-weighted images were compared with those obtained with the unenhanced images plus dynamic contrast material-enhanced single-section turbo fast low-angle shot (FLASH) images. Urinary bladder cancer started to enhance 6.5 seconds +/- 3.5 (standard deviation) after the beginning of arterial enhancement, which was 4 seconds earlier than most other structures (postbiopsy tissue, 13.6 seconds +/- 4.2). In differentiation of postbiopsy tissue from malignancy on the basis of the beginning of enhancement depicted on time and subtracted images, accuracy improved from 79% to 90% (P < .02) and specificity improved from 33% to 92% (not significant). Overall, tumor staging accuracy improved significantly from 67% to 84% (P < .01) by adding the turbo FLASH images. CONCLUSION: Fast dynamic first-pass MR imaging, with at least one image acquired every 2 seconds, improved delineation of urinary bladder cancer, tumor staging, and detection of metastases.

Biopsy↗
PubMed1993

Spatial resolution in digital mammography.

RATIONAL AND OBJECTIVES: Digital acquisition systems currently available limit spatial resolution in digital mammography to roughly 0.1 mm/pixel. The objective of this study is to determine if high-quality mammography is possible at this resolution. METHODS: The influence of spatial resolution on diagnostic quality was investigated by comparing observer performance on film to that on digitized film. A 0.1-mm sampling distance was used for digitization. Detection of mammographic details was studied by measuring threshold contrast as a function of detail size for small circular objects in the range of 0.12 to 2.5 mm. Characterization of microcalcifications was investigated in a receiver operating characteristic (ROC) study, in which 10 radiologists read 72 mammographic details with microcalcifications, both digitally and on film. RESULTS: Digitization improved the detectability of the larger, low contrast objects, whereas for small objects the detectability did not change. The authors found that even under the most optimal circumstances, isolated spherical calcifications with diameters smaller than 0.13 mm are not detectable with film-screen mammography, despite its resolution limit of 15 line patterns per mm (lp/mm). The ability to characterize microcalcification clusters did not change significantly with digitization. However, the results suggest that differentiation of benign from malignant cases decreases slightly, and that characterization of different types of malignancies somewhat improves by digitization. Mean differences between the two modalities were considerably smaller than the interobserver variability. CONCLUSION: A relatively low spatial resolution of 0.1 mm/pixel does not prohibit high-quality diagnostic performance in digital mammography.

Breast Diseases↗
PubMed1990

[Digital mammography is very useful in mass screening of breast cancer].

Mammograms made between 1981 and 1989 in the Nijmegen screening programme for breast cancer were retrospectively reviewed. Those made before detection of breast cancer showed signs of tumour growth in the place of the subsequently detected malignancy in 22% of the cases. A work station was set up for image digitization and image processing. Display with optimal contrast and image processing of mammograms is possible. Diagnoses based on digitized mammograms displayed on a monitor were as good as those based on the conventional images on film. Automatic detection of image features has been investigated. A procedure for automatic detection of microcalcifications was developed. This research is important for optimalization of diagnosis in screening and because of the expected introduction of direct digital imaging techniques.

Breast Neoplasms↗