Reproducibility and accuracy of interactive segmentation procedures for image analysis in cytology

Automatic segmentation of cell nuclei in Feulgen-stained histological sections of prostate cancer and quantitative evaluation of segmentation results

Digital image analysis of cell nuclei is useful to obtain quantitative information for the diagnosis and prognosis of cancer. However, the lack of a reliable automatic nuclear segmentation is a limiting factor for high-throughput nuclear image analysis. We have developed a method for automatic segmentation of nuclei in Feulgen-stained histological sections of prostate cancer. A local adaptive thresholding with an object perimeter gradient verification step detected the nuclei and was combined with an active contour model that featured an optimized initialization and worked within a restricted region to improve convergence of the segmentation of each nucleus. The method was tested on 30 randomly selected image frames from three cases, comparing the results from the automatic algorithm to a manual delineation of 924 nuclei. The automatic method segmented a few more nuclei compared to the manual method, and about 73% of the manually segmented nuclei were also segmented by the automatic method. For each nucleus segmented both manually and automatically, the accuracy (i.e., agreement with manual delineation) was estimated. The mean segmentation sensitivity/specificity were 95%/96%. The results from the automatic method were not significantly different from the ground truth provided by manual segmentation. This opens the possibility for large-scale nuclear analysis based on automatic segmentation of nuclei in Feulgen-stained histological sections.

Interactive measurement and characterization of DNA molecules by analysis of AFM images

BACKGROUND

In the past few years, computer-based analysis of atomic-force microscopic images has acquired increasing importance for studying biomolecules such as DNA. On the one hand, fully automated methods do not allow analysis of complex shapes; on the other hand, manual methods are usually time consuming and inaccurate. The semiautomated approach presented in this report overcomes the drawbacks of both methods.

METHODS

Two kinds of images were analyzed: computer-generated filaments that modeled circular DNA molecules on a surface and real atomic-force microscopic images of DNA molecules adsorbed on an appropriate substrate surface.

RESULTS

The algorithm was tested on a group of 140 simulated and 189 real plasmids with a nominal length of 913 nm. The accuracy of the length measurement was statistically evaluated on the ensemble of molecules, with particular attention to the influence of the noise. Mean contour lengths of 912 +/- 5 nm and 910 +/- 47 nm were found for simulated and real plasmids, respectively. The measured end-to-end distance of lambda-DNA molecules as a function of their contour length is reported, from which it is possible to estimate the stiffness of the DNA molecules adsorbed onto a surface; the value obtained for the DNA persistence length (42 +/- 5 nm) is consistent with values measured by other imaging techniques.

CONCLUSIONS

An interactive algorithm for DNA molecule measurements based on the detection of the filament ridge line in a digitized image is presented. The simulation of artificial filaments combined with the experimental data demonstrates that the proposed method can be a valuable tool for the DNA contour length evaluation, especially in the case of complex shapes where the use of automatic methods is not possible.

Some trends in microscope image processing

The present review tries to identify some trends among the multitude of ways followed by image processing developments in the field of microscopy. Nine topics were selected. They cover the fields of: signal processing, statistical analysis, artificial intelligence, three-dimensional microscopy, multidimensional microscopy, multimodality microscopy, theory, simulation and multidisciplinarity. A specific topic is dedicated to a trend towards semi-automation instead of full automation in image processing.

Quantitative image analysis in mammary gland biology

In this paper we present a summary of recent quantitative approaches used for the analysis of macro and microscopic images in mammary gland biology. The advantages and disadvantages of whole mount analysis, reconstruction of serial tissue sections and nucleus/cell segmentation of either conventional and confocal images are discussed, as are applications of quantitative image analysis, such as quantification of protein levels or vasculature measurements in normal tissue and cancer. Integration of quantitative imaging into the further study of the mammary gland holds the promise of better understanding its tissue complexity that evolves during development, differentiation and disease.

Interactive 3D editing tools for image segmentation

Segmentation is an important part of image processing, which often has a large impact on quantitative image analysis results. Fully automated operator independent segmentation procedures that successfully work in a population with a larger biological variation are extremely difficult to design and usually some kind of operator intervention is required, at least in pathological cases. We developed a variety of 3D editing tools that can be used to correct or improve results of initial automatic segmentation procedures. Specifically we will discuss and show examples for three types of editing tools that we termed: hole-filling (tool 1), point-bridging (tool 2), and surface-dragging (tool 3). Each tool comes in a number of flavors, all of which are implemented in a truly 3D manner. We describe the principles, evaluate efficiency and flexibility, and discuss advantages and disadvantages of each tool. We further demonstrate the superiority of the 3D approach over the time-consuming slice-by-slice editing of 3D datasets, which is still widely used in medical image processing today. We conclude that performance criteria for automatic segmentation algorithms may be eased significantly by including 3D editing tools early in the design process.

Applications of image analysis to anatomic pathology: realities and promises

Image Analysis in Pathology is viewed as an ancillary method meant to provide objective support in the resolution of difficult problems. Its Achilles heel is the process of nuclear segmentation (delimitation of the nuclear membrane) which is extremely difficult in pathology materials. Although interactive segmentation procedures are available no reliable fully automatic method has been described. The only application of image analysis that has truly succeeded in Pathology is DNA ploidy measurement. A very desirable application is the quantitation of immunohistochemical markers, which is technically challenging, has been resolved only in certain cases and is unlikely to have a general solution. Nuclear quantitation has repeatedly proven to be helpful in reaching differential diagnoses, in particular when based on size distributions of nuclear profiles rather than its average, but is hampered by the segmentation problem discussed above. Texture analysis of chromatin is an exciting, mathematically complex application likely to succeed, for which many approaches have been described. Finally a diagnosis (classification) can be obtained based on algorithms applied to multiple descriptors of tumor cells (for instance nuclear sizes, chromatin texture, shape, etc). The best classificatory approaches are neural networks (a form of artificial intelligencee), multivariate analysis, and logistic regression (statistical).

Human papillomavirus-related cervical lesions in adolescents: a histologic and morphometric study

OBJECTIVE

To evaluate the cytopathic effect of human papillomavirus (HPV) infection in adolescents.

METHODS

Cervical biopsies from 100 patients, 50 from adolescents age 14 to 20 and 50 from mature women age 35 to 64, all diagnosed with HPV-related lesions (condylomas), were studied histologically and morphometrically. Fifty were associated with low-grade squamous intraepithelial lesions and 50 with high-grade squamous intraepithelial lesions.

RESULTS

Epithelial cells with large hyperchromatic, often bizarre-shaped, nuclei, staining positive for HPV-16, were identified in most samples from adolescents. These nuclei were, on average, almost twice as large as those seen in biopsies from older women (P = 0.003), and they mostly occupied the lower half of the epithelium (P = 0.0008). These large cells were further analyzed for texture by a novel image-analysis approach, the autocorrelation factor beta, that revealed a markedly different, smoother nuclear structure, suggestive of a large viral load, different from the irregular chromatin pattern seen in dysplastic nuclei.

CONCLUSIONS

These peculiar nuclei were ubiquitous in the adolescent cervical biopsies and probably represent a primary abundant productive viral infection. They should not necessarily be interpreted as dysplastic.

Image analysis and morphometry in the diagnosis of breast cancer

Image Analysis, a complicated field still in the early stages of application to Pathology, has the capability of rendering major contributions to the diagnosis, prognosis, and management of malignancies of the breast. The present review summarizes the main problems and the general approach to the use of this technique for quantitating immunohistochemical stain results, obtaining DNA histograms, and making de novo diagnoses in routine materials of the Pathology service. In the case of diagnosis, the main steps are sampling, segmentation, and measures of chromatin texture. Currently, the limiting factor for all routine applications of image analysis is probably the absence of a reliable automatic nuclear segmentation.

Interactive 3-dimensional segmentation of MRI data in personal computer environment

We describe a method of interactive three-dimensional segmentation and visualization for anatomical magnetic resonance imaging (MRI) data in a personal computer environment. The visual feedback necessary during 3-D segmentation was provided by a ray casting algorithm, which was designed to allow users to interactively decide the visualization quality depending on the task-requirement. Structures such as gray matter, white matter, and facial skin from T1-weighted high-resolution MRI data were segmented and later visualized with surface rendering. Personal computers with central processing unit (CPU) speeds of 266, 400, and 700 MHz, were used for the implementation. The 3-D visualization upon each execution of the segmentation operation was achieved in the order of 2 s with a 700 MHz CPU. Our results suggest that 3-D volume segmentation with semi real-time visual feedback could be effectively implemented in a PC environment without the need for dedicated graphics processing hardware.

Fractal characterization of chromatin appearance for diagnosis in breast cytology

This study explores the use of fractal analysis in the numerical description of chromatin appearance in breast cytology. Images of nuclei from fine-needle aspiration biopsies of the breast are characterized in terms of their Minkowski and spectral fractal dimensions, for 19 patients with benign epithelial cell lesions and 22 with invasive ductal carcinomas. Chromatin appearance in breast epithelial cell nuclear images is demonstrated to be fractal, suggesting that the three-dimensional chromatin structure in these cells also has fractal properties. A statistically significant difference between the mean spectral dimensions of the benign and malignant cases is demonstrated. The two fractal dimensions are very weakly correlated. A statistically significant difference between the benign and malignant cases in lacunarity, a fractal property characterizing the size of holes or gaps in a texture, is found over a wide range of scales. These differences are particularly pronounced at the smallest and largest scales, corresponding respectively to fine-scale texture, indicating whether chromatin is clumped or fine, and to large-scale structures like nucleoli. Logistic regression and artificial neural network classification models are developed to classify unknown cases on the basis of fractal measures of chromatin texture. Using leave-one-out cross-validation, the best logistic regression classifier correctly diagnoses 95.1 per cent of the cases. The best neural network model can correctly classify all of the cases, but it is unclear whether this is due to overtraining. Fractal dimensions and lacunarity are useful tools for the quantitative characterization of chromatin appearance, and can potentially be incorporated into image analysis devices to assure the quality and reproducibility of diagnosis by breast fine-needle aspiration biopsy.