Birgitte Nielsen

Adaptive gray level run length features from class distance matrices

We constructed class distance matrices for the gray level run length texture analysis method. For a four-class problem of liver cell nuclei, we found that there exist areas of consistently high values in the class distance matrices. We combined the information from the entries of the normalized run length matrix, based on the class distance matrices, to obtain adaptive features for texture classification. Using this procedure, we extracted only two features, which halved the classification error when compared to the best pair of classical gray level run length matrix features.

The use of fractal features from the periphery of cell nuclei as a classification tool.

A polygonization‐based method is used to estimate the fractal dimension and several new scalar lacunarity features from digitized transmission electron micrographs (TEM) of mouse liver cell nuclei. The fractal features have been estimated in different segments of 1D curves obtained by scanning the 2D cell nuclei in a spiral‐like fashion called “peel‐off scanning”. This is a venue to separate estimates of fractal features in the center and periphery of a cell nucleus. Our aim was to see if a small set of fractal features could discriminate between samples from normal liver, hyperplastic nodules and hepatocellular carcinomas. The Bhattacharyya distance was used to evaluate the features. Bayesian classification with pooled covariance matrix and equal prior probabilities was used as the rule for classification. Several single fractal features estimated from the periphery of the cell nuclei discriminated samples from the hyperplastic nodules and hepatocellular carcinomas from normal ones. The outer 25–30% of the cell nuclei contained important texture information about the differences between the classes. The polygonization‐based method was also used as an analysis tool to relate the differences between the classes to differences in the chromatin structure.

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.

Comparison of nuclear texture analysis and image cytometric DNA analysis for the assessment of dysplasia in Barrett's oesophagus

Background:Dysplasia is a marker of cancer risk in Barretts oesophagus (BO), but this risk is variable and diagnosis is subject to inter-observer variability. Cancer risk in BO is increased when chromosomal instability is present. Nucleotyping (NT) is a new method that uses high-resolution digital images of nuclei to assess chromatin organisation both quantitatively and qualitatively. We aimed to evaluate NT as a marker of dysplasia in BO and compare with image cytometric DNA analysis (ICM).Methods:In all, 120 patients with BO were studied. The non-dysplastic group (n=60) had specialised intestinal metaplasia only on two consecutive endoscopies after 51 months median follow-up (IQR=25–120 months). The dysplastic group (n=60) had high-grade dysplasia or carcinoma in situ. The two groups were then randomly assigned to a training set and a blinded test set in a 1 : 1 ratio. Image cytometric DNA analysis and NT was then carried out on Feulgen-stained nuclear monolayers.Results:The best-fit model for NT gave a correct classification rate (CCR) for the training set of 83%. The test set was then analysed using the same textural features and yielded a CCR of 78%. Image cytometric DNA analysis alone yielded a CCR of 73%. The combination of ICM and NT yielded a CCR of 84%.Conclusion:Nucleotyping differentiates dysplastic and non-dysplastic BO, with a greater sensitivity than ICM. A combination score based on both techniques performed better than either test in isolation. These data demonstrate that NT/ICM on nuclear monolayers is a very promising single platform test of genomic instability, which may aid pathologists in the diagnosis of dysplasia and has potential as a biomarker in BO.

Fractal Analysis of Monolayer Cell Nuclei from Two Different Prognostic Classes of Early Ovarian Cancer

Most women undergoing treatment for early ovarian cancer have a good prognosis, but about 20% will eventually die of the disease. Identifying patients with increased risk of relapse is important, as it could be used to select patients in need for adjuvant treatment after surgery. The aim of the present study has been to analyze the prognostic value of nuclear fractal features in early ovarian cancer, and to study the complex relation between nuclear area, nuclear DNA content, nuclear gray level distribution and nuclear fractal features. We found that the monolayer nuclei from a given lesion differed widely in fractal dimension. The fractal dimension in the peripheral part of the nuclei was higher than the fractal dimension in the central part of the nuclei. The intra-patient variability of fractal dimension was larger than the inter-patient variability of the mean fractal dimension. Fractal dimension was insufficient for classification. The cell nuclei were grouped into area bins according to nuclear area. Lacunarity class distance and class difference matrices were extracted from the nuclei within each area bin. Some few area intervals contained most of the class distance information between the two prognostic classes of early ovarian cancer. The Mahalanobis values contained in the class distance matrices computed from these area bins were about four times higher than the Mahalanobis values contained in the area independent class distance matrices computed from all the nuclei. However, the lacunarity features were not sufficient to discriminate the two classes of early ovarian cancer.

Prognostic classification of early ovarian cancer based on very low dimensionality adaptive texture feature vectors from cell nuclei from monolayers and histological sections.

In order to study the prognostic value of quantifying the chromatin structure of cell nuclei from patients with early ovarian cancer, low dimensionality adaptive fractal and Gray Level Cooccurrence Matrix texture feature vectors were extracted from nuclei images of monolayers and histological sections. Each light microscopy nucleus image was divided into a peripheral and a central part, representing 30% and 70% of the total area of the nucleus, respectively. Textural features were then extracted from the peripheral and central parts of the nuclei images. The adaptive feature extraction was based on Class Difference Matrices and Class Distance Matrices. These matrices were useful to illustrate the difference in chromatin texture between the good and bad prognosis classes of ovarian samples. Class Difference and Distance Matrices also clearly illustrated the difference in texture between the peripheral and central parts of cell nuclei. Both when working with nuclei images from monolayers and from histological sections it seems useful to extract separate features from the peripheral and central parts of the nuclei images.

Entropy-based adaptive nuclear texture features are independent prognostic markers in a total population of uterine sarcomas

Nuclear texture analysis measures the spatial arrangement of the pixel gray levels in a digitized microscopic nuclear image and is a promising quantitative tool for prognosis of cancer. The aim of this study was to evaluate the prognostic value of entropy‐based adaptive nuclear texture features in a total population of 354 uterine sarcomas. Isolated nuclei (monolayers) were prepared from 50 µm tissue sections and stained with Feulgen‐Schiff. Local gray level entropy was measured within small windows of each nuclear image and stored in gray level entropy matrices, and two superior adaptive texture features were calculated from each matrix. The 5‐year crude survival was significantly higher (P < 0.001) for patients with high texture feature values (72%) than for patients with low feature values (36%). When combining DNA ploidy classification (diploid/nondiploid) and texture (high/low feature value), the patients could be stratified into three risk groups with 5‐year crude survival of 77, 57, and 34% (Hazard Ratios (HR) of 1, 2.3, and 4.1, P < 0.001). Entropy‐based adaptive nuclear texture was an independent prognostic marker for crude survival in multivariate analysis including relevant clinicopathological features (HR = 2.1, P = 0.001), and should therefore be considered as a potential prognostic marker in uterine sarcomas.

Prognostic Value of Adaptive Textural Features–The Effect of Standardizing Nuclear First-Order Gray Level Statistics and Mixing Information from Nuclei Having Different Area

Background: Nuclear texture analysis is a useful method to obtain quantitative information for use in prognosis of cancer. The first-order gray level statistics of a digitized light microscopic nuclear image may be influenced by variations in the image input conditions. Therefore, we have previously standardized the nuclear gray level mean value and standard deviation. However, there is a clear relation between nuclear DNA content, area, first-order statistics, and texture. For nuclei with approximately the same DNA content, the mean gray level increases with an increasing nuclear area. The aims of the present methodical work were to study: (1) whether the prognostic value of adaptive textural features varies with nuclear area, and (2) the effect of standardizing nuclear first-order statistics. Methods: Nuclei from 134 cases of ovarian cancer were grouped into intervals according to nuclear area. Adaptive features were extracted from two different image sets, i.e., standardized and non-standardized nuclear images. Results: The prognostic value of adaptive textural features varied strongly with nuclear area. A standardization of the first-order statistics significantly reduced this prognostic information. Several single features discriminated the two classes of cancer with a correct classification rate of 70%. Conclusion: Nuclei having an area between 2000–4999 pixels contained most of the class distance information between the good and poor prognosis classes of cancer. By considering the relation between nuclear area and texture, we avoided a loss of information caused by standardizing the first-order statistics and mixing data from cells having different nuclear area.

The prognostic value of adaptive nuclear texture features from patient gray level entropy matrices in early stage ovarian cancer

Background: Nuclear texture analysis gives information about the spatial arrangement of the pixel gray levels in a digitized microscopic nuclear image, providing texture features that may be used as quantitative tools for prognosis of human cancer. The aim of the study was to evaluate the prognostic value of adaptive nuclear texture features in early stage ovarian cancer. Methods: 246 cases of early stage ovarian cancer were included in the analysis. Isolated nuclei (monolayers) were prepared from 50 μm tissue sections and stained with Feulgen-Schiff. Local gray level entropy was measured within small windows of each nuclear image and stored in gray level entropy matrices. A compact set of adaptive features was computed from these matrices. Results: Univariate Kaplan-Meier analysis showed significantly better relapse-free survival (p < 0.001) for patients with low adaptive feature values compared to patients with high adaptive feature values. The 10-year relapse-free survival was about 78% for patients with low feature values and about 52% for patients with high feature values. Adaptive features were found to be of independent prognostic significance for relapse-free survival in a multivariate analysis. Conclusion: Adaptive nuclear texture features from entropy matrices contain prognostic information and are of independent prognostic significance for relapse-free survival in early stage ovarian cancer.

Fractal Signature Vectors and Lacunarity Class Distance Matrices to Extract New Adaptive Texture Features from Cell Nuclei

A polygonization-based method is used to compute Fractal Signature Vectors and Lacunarity Matrices from digitized transmission electron micrographs of liver cell nuclei from normal and premalignant samples. Based on all the cells in each of the two classes we have constructed a Fractal Signature Class Distance Vector and a Lacunarity Class Distance Matrix. We have found that there exist areas of consistently high values in the class distance vector (matrix). Adaptive fractal signature and lacunarity features are obtained by weighted summations of the fractal signature vectors and lacunarity matrices, using the class distance vector (matrix) as weights. Such features are adaptive to the classification problem at hand. The feature extraction of four new adaptive fractal features could replace the extraction of seven pre-defined fractal dimension and lacunarity features.