Multispectral Intensity, Textural & Morphology-based Mitosis detection in Multispectral images (MITM3) Framework

Get Complete Project Material File(s) Now! »

Color Normalization

Many color normalization techniques have been proposed [64, 112, 109, 93], including histogram or quantile normalization in which the distributions of the three color channels are normalized separately. Kothari et al [93] used histogram based normalization in histopathological images. They proposed a rank function that maps the intensity ranges across all pixels. Alternatively, Reinhard et al [143] proposed a method for matching the color distribution of an image to that of reference image by use of a linear transform in a perceptual color model (Lab). Magee et al [111] extended Reinhard’s normalization approach to multiple pixel classes by using a probabilistic (GMM) color segmentation method. It applies a separate linear normalization for each pixel where class membership is defined by a pixel being coloured by a particular chemical stain or being uncoloured i.e., background. In order to deal with stains colocalization, a very common phenomenon in histopathological images, color deconvolution is effective in separation of stains [153]. Ruifrok [153] explains how virtually every set of three colors can be separated by color deconvolution and reconstructs for each stain separately. It requires prior knowledge of color vectors (RGB) of each specific stain. Later, Macenko et al [109] proposed the automatic derivation of these color vectors, a method further refined by Niethammer et al [129] and Magee et al [111].
Several nuclei detection and segmentation methods [37, 27, 175, 173, 87] are using color deconvolution based separation of stains in the histopathological images. Different color models can be used. Several detection and segmentation methods [67, 68, 39, 74, 27, 8, 140] use the RGB color model, yet the RGB model is not a perceptually uniform color model. Other more perceptual color models such as HSV, Lab and Luv are sometimes used [185, 186, 17, 13, 45, 128, 91, 87, 88, 113].

Noise Reduction and Image Smoothing

Thresholding is used for noise reduction that usually follows filtering and background correction in order to minimize random noise and artefacts [15, 85]. The pixels that lie outside threshold values often determined using intensity histogram are considered to be noisy. Alternatively, applying threshold function on a group of pixels instead of an individual pixel eliminates a noisy region. While such techniques are successful to eliminate small spots of noise, they fail at eliminating large artefacts [69].
Alternatively, morphological operations can also be used for noise reduction. Noise and artefacts are eliminated using morphological operations like closings and openings [175]. Morphological grayscale reconstruction methods are used to eliminate noise while preserving the nuclei shape [74, 79, 80, 88]. While thresholding and filtering reduce noise according to pixel intensities, morphology reduces noise based on the shape characteristics of the input image, as characterized by a structuring element. Morphology cannot distinguish the cellular areas and artefacts having a cell-like shape but different intensity values. Thresholding (prior or subsequent to applying the morphological operations) removes such artefacts. Adaptive filters [62], Gamma correction [39], and histogram equalization [157] have been used to increase the contrast between foreground (nuclei) and background regions. Anisotropic diffusion is used to smooth nuclei information without degrading nuclei edges [157, 87]. Gaussian filtering is also used to smooth nuclei regions [177, 17, 124].

Region Of Interest Detection

In some frameworks, noise reduction and ROI detection are performed at the same time. For instance, in the case of tissue-level feature extraction, the pre-processing step thresholds the image to identify the ROI by eliminating both noisy regions and those with little content [69]. While, in case of cellular-level feature extraction, noise reduction is followed by ROI detection to determine the nuclei region [87, 88]. Thresholding is popular for ROI detection. In follicular lymphoma (FL) tissue, there are five cytological components: nuclei, cytoplasm, extra-cellular material, red blood cells (RBCs) and background regions. Sertel et al [157] introduced the nuclei and cytological components as ROI for grading of FL. RBCs and background regions show uniform patterns as compared to other nuclei in FL tissue; thus thresholding is performed in RGB color model for elimination of RBCs and background. Similarly, Dalle et al [39] selected neoplasm ROI by using Otsu thresholding along with morphological operations. Clustering is another method that commonly used for ROI detection. Cataldo et al [27] performed the automated separation of cancer from non-cancerous regions (stroma, blood vessels) using unsupervised clustering. Later, cancer and non-cancerous regions is refined using morphological operations. Dundar et al [45] proposed a framework for classification of intraductal breast lesions as benign or malignant using cellular component. The intraductal breast lesions contain four components: cellular, extra cellular, regions with hues of red and illumina. The H&E stained image data is modeled into four components using GMM. Parameters of GMM model are estimated using EM [43]. The resulting mixture distribution is used to classify pixels into four categories. Those classified as cellular component are further clustered by dynamic thresholding to eliminate blue-purple pixels with relatively less luminance. The remaining pixels are considered cellular region and is used in lesion classification.
Using textural information, Khan et al [88] proposed a novel and unsupervised approach to segment breast cancer histopathology images into two regions; Hypo-Cellular Stroma (HypoCS) and Hyper-Cellular Stroma (HyperCS). This approach is employed magnitude and phase spectrum in the Gabor frequency domain to segment HypoCS and HyperCS regions, respectively. For mitosis detection in breast cancer histopathology images using this approach as ROI detection, it reduces the false positive rate (FPR) from four times [87].

READ  Electrochemistry in Lead Dioxide Electrodeposition

Nuclei Detection, Segmentation and ClassificationMethods

Identification of initial markers or seed points, usually one per nuclei and close to its center, is a pre-requisite for most nuclei segmentation methods. The subsequent frameworks use seeds points in order to delineate the spatial extent of each nuclei. Indeed, the accuracy of such segmentation methods depends critically on the reliability of the seed points. The early works in this field relies upon the peaks of the Euclidean distance map [39]. H-maxima transform detects local maxima as seed points [177, 166, 79, 80] but it is overly sensitive to texture and often results in overseeding. Hough transform detects seed points for circular shaped nuclei but requires heavy computation [37]. Centroid transform also detects seeds but limitations make it useful only for binarized images and unable to exploit additional cues.
The Euclidean distance map is commonly used for seeds detection and Laplacian of Gaussian (LoG) is a generic blob detection method. Using multiscale LoG filter with a Euclidean distance map offers important advantages, including computational efficiency and ability to exploit shape and sizes information. Al-kofahi et al [9] proposed a distance constrained multiscale LoG filtering method to identify the center of nuclei by exploiting shape and size cues available in the Euclidean distance map of the binarized image. The main steps of this methodology as follow:
i. Initially, compute the response of the scale-normalized LoG filter (LoGnorm(i; ξ) = ξ2 LoG(i; ξ)) at multiple scales ξ = [ξmin, · · · , ξmax].
ii. Use the Euclidean distance map DN(i) to constrain the maximum scale values when combining the LoG filtering results across scales to compute a single response surface RN(i) as: RN(i) = arg max ξ∈[ξmin,ξMAX] {LoGnorm(i; ξ) ∗ IN(i)}.

Table of contents :

List of Figures
List of Tables
1 Role of Image Analysis in Histopathology 
1.1 Introduction
1.2 Histopathology
1.3 Histopathology Imaging
1.4 Computer Aided Diagnosis Systems in Histopathology
1.5 Cancer and Grading System
1.6 Motivation of Our Study
1.7 Thesis Strucure
1.8 Conclusion
2 Review of Quantitative Image Analysis Methods in Histopathology 
2.1 Introduction
2.2 Image-Processing Methods
2.3 Preprocessing
2.4 Nuclei Detection, Segmentation and Classification Methods
2.5 Spectral and Spatial Characterization
2.6 Performance Metrics
2.7 Evaluation Methods
2.8 Inspection and Editing Software
2.9 Limitations and Challenges in Previous Frameworks
2.10 Overview of Proposed Framework and Scientific Contributions
2.11 Conclusion
3 Automated Mitosis Detection in Color (RGB) Images 
3.1 Introduction
3.2 Challenges in Mitosis Count
3.3 Color Dataset
3.4 Textural based Mitosis detection in Color images (TMC) Framework
3.5 Intensity, Textural and Morphology based Mitosis detection in Color images (ITM2C) Framework
3.6 MICO Platform Prototype
3.7 Conclusion
4 Automated Mitosis Detection in Multispectral Images 
4.1 Introduction
4.2 Multispectral Dataset
4.3 Multispectral Intensity, Textural & Morphology-based Mitosis detection in Multispectral images (MITM3) Framework
4.4 Experiments and Results
4.5 Discussion
4.6 Conclusion
5 Orientable 2 – Manifold Meshes and Dynamic Sampling 
5.1 Introduction
5.2 Surfaces and Meshes
5.3 Duality
5.4 Implement Duality in ITK
5.5 Validation
5.6 Dynamic Sampling for Cyto-Nuclear Atypia Score in MICO Platform
5.7 Conclusion
6 Overall Conclusion and Future Perspectives 
6.1 Conclusion
6.2 Future Perspectives
A Glossary


Related Posts