Andrew Mehnert

An improved seeded region growing algorithm

Recently Adams and Bischof (1994) proposed a novel region growing algorithm for segmenting intensity images. The inputs to the algorithm are the intensity image and a set of seeds - individual points or connected components - that identify the individual regions to be segmented. The algorithm grows these seed regions until all of the image pixels have been assimilated. Unfortunately the algorithm is inherently dependent on the order of pixel processing. This means, for example, that raster order processing and anti-raster order processing do not, in general, lead to the same tessellation. In this paper we propose an improved seeded region growing algorithm that retains the advantages of the Adams and Bischof algorithm - fast execution, robust segmentation, and no tuning parameters - but is pixel order independent.

Denoising of Dynamic Contrast-Enhanced MR Images Using Dynamic Nonlocal Means

This paper presents a new algorithm for denoising dynamic contrast-enhanced (DCE) MR images. It is a novel variation on the nonlocal means (NLM) algorithm. The algorithm, called dynamic nonlocal means (DNLM), exploits the redundancy of information in the temporal sequence of images. Empirical evaluations of the performance of the DNLM algorithm relative to seven other denoising methods-simple Gaussian filtering, the original NLM algorithm, a trivial extension of NLM to include the temporal dimension, bilateral filtering, anisotropic diffusion filtering, wavelet adaptive multiscale products threshold, and traditional wavelet thresholding-are presented. The evaluations include quantitative evaluations using simulated data and real data (20 DCE-MRI data sets from routine clinical breast MRI examinations) as well as qualitative evaluations using the same real data (24 observers: 14 image/signal-processing specialists, 10 clinical breast MRI radiographers). The results of the quantitative evaluation using the simulated data show that the DNLM algorithm consistently yields the smallest MSE between the denoised image and its corresponding original noiseless version. The results of the quantitative evaluation using the real data provide evidence, at the ¿ = 0.05 level of significance, that the DNLM algorithm yields the smallest MSE between the denoised image and its corresponding original noiseless version. The results of the qualitative evaluation provide evidence, at the ¿ = 0.05 level of significance, that the DNLM algorithm performs visually better than all of the other algorithms. Collectively the qualitative and quantitative results suggest that the DNLM algorithm more effectively attenuates noise in DCE MR images than any of the other algorithms.

Synthesizing dental radiographs for human identification.

The task of identifying human remains based on dental comparisons of post mortem (PM) and ante mortem (AM) radiographs is labor-intensive, subjective, and has several drawbacks, including: inherently poor image quality, difficulty matching the viewing angles in PM radiographs to those taken AM, and the fact that the state of the dental remains may entirely preclude the possibility of obtaining certain types of radiographs PM. The aim of the present study was to investigate the feasibility of using radiograph-like images reconstructed from PM x-ray computed tomography (CT) data to overcome the shortcomings of conventional radiographic comparison. Algorithms for computer synthesis of panoramic, periapical, and bitewing images are presented. The algorithms were evaluated with data from clinical examinations of two persons. The results demonstrate the efficacy of the CT-based approach and that, in comparison with conventional radiographs, the synthesized images exhibit minimal geometric distortion, reduced blurring, and reduced superimposition of oral structures.

Dental CT metal artefact reduction based on sequential substitution

OBJECTIVE Metal artefacts can seriously degrade the visual quality and interpretability of dental CT images. Existing image processing algorithms for metal artefact reduction (MAR) are either too computationally expensive to be used in clinical scanners or effective only in correcting mild artefacts. The aim of the present study was to investigate whether it is possible to improve the efficacy of the computationally efficient projection-correction approach to MAR by exploiting the spatial dependency or autocorrelation between adjacent CT slices. METHODS A new projection-correction algorithm [MAR by sequential substitution (MARSS)] was developed based on the idea that the corrupted portions of the projection data can be substituted with the corresponding portions from an unaffected adjacent slice. The performance of MARSS was evaluated relative to the projection-correction method of Watzke and Kalendar using a two-alternative forced choice (2AFC) visual trial involving 20 observers and 20 clinical CT data sets.16 RESULTS The Cochran Q test revealed no significant difference in the responses across all observers. The data were then pooled and analysed using a one-tailed exact binomial test. This revealed that the proportion of responses in favour of MARSS was significant (P < 2.2 × 10(-16)). A second Cochran Q test revealed no significant difference in the responses across all images. CONCLUSIONS It is possible to improve the efficacy of projection correction by exploiting spatial autocorrelation. The 2AFC results suggest that the proposed MARSS algorithm outperforms competing computationally efficient algorithms in terms of reducing metal artefacts whilst at the same time preserving/revealing anatomic detail.

On Computing the Exact Euclidean Distance Transform on Rectangular and Hexagonal Grids

In this paper we prove an equivalence relation between the distance transform of a binary image, where the underlying distance is based on a positive definite quadratic form, and the erosion of its characteristic function by an elliptic poweroid structuring element. The algorithms devised by Shih and Mitchell [18] and Huang and Mitchell [7], for calculating the exact Euclidean distance transform (EDT) of a binary digital image manifested on a square grid, are particular cases of this result. The former algorithm uses erosion by a circular cone to calculate the EDT whilst the latter uses erosion by an elliptic paraboloid (which allows for pixel aspect ratio correction) to calculate the square of the EDT. Huang and Mitchells algorithm [7] is arguably the better of the two because: (i) the structuring element can be decomposed into a sequence of dilations by 3 × 3 structuring elements (a similar decomposition is not possible for the circular cone) thus reducing the complexity of the erosion, and (ii) the algorithm only requires integer arithmetic (it produces squared distance). The algorithm is amenable to both hardware implementation using a pipeline architecture and efficient implementation on serial machines. Unfortunately the algorithm does not directly transpose to, nor has a corresponding analogue on, the hexagonal grid (the same is also true for Shih and Mitchells algorithm [7]). In this paper, however, we show that if the hexagonal grid image is embedded in a rectangular grid then Huang and Mitchells algorithm [7] can be applied, with aspect ratio correction, to obtain the exact EDT on the hexagonal grid.

An evaluation of four parametric models of contrast enhancement for dynamic magnetic resonance imaging of the breast

This paper presents an empirical evaluation of the goodness-of-fit (GOF) of four parametric models of contrast enhancement for dynamic resonance imaging of the breast: the Tofts, Brix, and Hayton pharmacokinetic models, and a novel empiric model. The goodness-of-fit of each model was evaluated with respect to: (i) two model-fitting algorithms (Levenberg- Marquardt and Nelder-Mead) and two fitting tolerances; and (ii) temporal resolution. In the first case the GOF was measured using data from three dynamic contrast-enhanced (DCE) MRI data sets from routine clinical examinations: one case with benign enhancement, one with malignant enhancement, and one with normal findings. Results are presented for fits to both the whole breast volume and to a selected region of interest. In the second case the GOF was measured by first fitting the models to several temporally sub-sampled versions of a custom high temporal resolution data set (subset of the breast volume containing a malignant lesion), and then comparing the fitted results to the original full temporal resolution data. Our results demonstrate that under the various optimization conditions considered, in general, both the proposed empiric model and the Hayton model fit the data equally well and that both of these models fit the data better than the Tofts and Brix models.

Synthesizing panoramic radiographs by unwrapping dental CT data

A method for synthesizing panoramic radiographs from dental CT data is presented. The method is based on the principles of panoramic radiography with a continuously-moving rotation center. The method computes discrete pixel sums through the CT data along normals to the medial axis of the dental arch. Compared to a conventional panoramic radiograph, the method produces less geometric distortion, less blurring, and less superimposition of other structures. The method is particularly suited to forensic identification of human remains in cases where the state of degradation precludes the possibility of obtaining a conventional panoramic radiograph

New Spatiotemporal Features for Improved Discrimination of Benign and Malignant Lesions in Dynamic Contrast-Enhanced-Magnetic Resonance Imaging of the Breast

Objectives: The objective of this study was to measure the efficacy of 7 new spatiotemporal features for discriminating between benign and malignant lesions in dynamic contrast-enhanced-magnetic resonance imaging (MRI) of the breast. Methods: A total of 48 breast lesions from 39 patients were used: 25 malignant and 23 benign. Lesions were acquired using 1.5-T MRI machines in 3 different protocols. Two experiments were performed: (i) selection of the most discriminatory subset of features drawn from the new features and features from the literature and (ii) validation of classification performance of the selected subset of features. Results: Results of the feature selection experiment show that the subset comprising 2 of the new features is the most useful for automatic classification of suspicious lesions in the breast: (i) gradient correlation of maximum intensity and (ii) mean wash-in rate. Results of the validation experiment show that using these 2 features, unseen data can be classified with an area under the receiver operating characteristic curve of 0.91 ± 0.06. Conclusions: Results of the experiments suggest that suspicious lesions in dynamic contrast-enhanced-MRI of the breast can be classified, with high accuracy, using only 2 of the proposed spatiotemporal features. The selected features indicate heterogeneity of enhancement and speed of enhancement in a tissue. High values of these indicators are likely to be correlated with malignancy.

Dynamic breast MRI: Image registration and its impact on enhancement curve estimation

A novel algorithm for performing registration of dynamic contrast-enhanced (DCE) MRI data of the breast is presented. It is based on an algorithm known as iterated dynamic programming originally devised to solve the stereo matching problem. Using artificially distorted DCE-MRI breast images it is shown that the proposed algorithm is able to correct for movement and distortions over a larger range than is likely to occur during routine clinical examination. In addition, using a clinical DCE-MRI data set with an expertly labeled suspicious region, it is shown that the proposed algorithm significantly reduces the variability of the enhancement curves at the pixel level yielding more pronounced uptake and washout phases

Novel chromatin texture features for the classification of pap smears

This paper presents a set of novel structural texture features for quantifying nuclear chromatin patterns in cells on a conventional Pap smear. The features are derived from an initial segmentation of the chromatin into bloblike texture primitives. The results of a comprehensive feature selection experiment, including the set of proposed structural texture features and a range of different cytology features drawn from the literature, show that two of the four top ranking features are structural texture features. They also show that a combination of structural and conventional features yields a classification performance of 0.954±0.019 (AUC±SE) for the discrimination of normal (NILM) and abnormal (LSIL and HSIL) slides. The results of a second classification experiment, using only normal-appearing cells from both normal and abnormal slides, demonstrates that a single structural texture feature measuring chromatin margination yields a classification performance of 0.815±0.019. Overall the results demonstrate the efficacy of the proposed structural approach and that it is possible to detect malignancy associated changes (MACs) in Papanicoloau stain.