Ilige S. Hage

Segmentation of histology slides of cortical bone using pulse coupled neural networks optimized by particle-swarm optimization

The aim of this study is to automatically discern the micro-features in histology slides of cortical bone using pulse coupled neural networks (PCNN). To the best knowledge of the authors, utilizing PCNN in such an application has not been reported in the literature and, as such, constitutes a novel application. The network parameters are optimized using particle swarm optimization (PSO) where the PSO fitness function was introduced as the entropy and energy of the bone micro-constituents extracted from a training image. Another novel contribution is combining the above with the method of adaptive threshold (T) where the PCNN algorithm is repeated until the best threshold T is found corresponding to the maximum variance between two segmented regions. To illustrate the quality of resulting segmentation according to this methodology, a comparison of the entropy/energy obtained of each pulse is reported. Suitable quality metrics (precision rate, sensitivity, specificity, accuracy, and dice) were used to benchmark the resulting segments against those found by a more traditional method namely K-means. The quality of the segments revealed by this methodology was found to be of much superior quality. Another testament to the quality of this methodology was that the images resulting from testing pulses were found to be of similarly good quality to those of the training images.

Structural Micro Processing of Haversian Systems of a Cortical Bovine Femur Using Optical Stereomicroscope and MATLAB

Processing of optical images of bone has been a topic of considerable interest in the past and continues to be so. Image processing can be used in medicine in order to improve the image visualization to detect diseases, and to compute properties such as area for abnormal cells. Several studies of bone images have been conducted using several methods including segmentation and image enhancement. The aim of this paper is to generate a standalone automated code for segmenting colored optical microscope images in order to show the microstructure of a cortical bone as a multi-phase (here 4 phases) composite: Lamella (matrix), Haversian canals, osteoblast lamella boundaries (freshly generated lamella lining), and lacunae (containing living cells).For this purpose, we investigate the use of MATLAB, which contains image-processing toolboxes with many analytical capabilities that have been advertised to be useful for many applications including biological systems. In this work, such capabilities are utilized in image processing of the microstructure of bovine cortical bone, which is generally accepted as proxy for human bone. Two specimens of the cortical regions of a bovine femur bones were imaged using Olympus optical microscope. One of the specimens was treated with the Masson’s trichrome staining treatment and the other with the Hematoxylin and Eosin (H&E) treatment. The images from the microscope were captured using a DP12 camera.Furthermore, MATLAB results are contrasted against Stream®, a commercially available software package procured along with the Olympus optical microscope. Via color-coding to facilitate the bone microstructure identification, the image analysis results were compared after computing the areas of each of the 4 constituent microstructural phases. Areas of each phase were calculated and comparisons made between the results obtained from the Stream® software and those obtained from MATLAB. The relative error was found to be quite small (<1%), which proves that MATLAB may be an effective software for medical image processing and may be the tool of choice for standalone applications.Copyright

Geometric-attributes-based segmentation of cortical bone slides using optimized neural networks.

In cortical bone, solid (lamellar and interstitial) matrix occupies space left over by porous microfeatures such as Haversian canals, lacunae, and canaliculi-containing clusters. In this work, pulse-coupled neural networks (PCNN) were used to automatically distinguish the microfeatures present in histology slides of cortical bone. The networks’ parameters were optimized using particle swarm optimization (PSO). When forming the fitness functions for the PSO, we considered the microfeatures’ geometric attributes—namely, their size (based on measures of elliptical perimeter or area), shape (based on measures of compactness or the ratio of minor axis length to major axis length), and a two-way combination of these two geometric attributes. This hybrid PCNN–PSO method was further enhanced for pulse evaluation by combination with yet another method, adaptive threshold (AT), where the PCNN algorithm is repeated until the best threshold is found corresponding to the maximum variance between two segmented regions. Together, this framework of using PCNN–PSO–AT constitutes, we believe, a novel framework in biomedical imaging. Using this framework and extracting microfeatures from only one training image, we successfully extracted microfeatures from other test images. The high fidelity of all resultant segments was established using quantitative metrics such as precision, specificity, and Dice indices.

Micro FEM Simulations of Single-Cutting-Edge Sawing of Cortical Bone

The finite element method at the micro scale (mFEM) has been gaining in popularity to simulate biomechanical effects. In this paper, a 3D mFEM model is developed to simulate sawing of cortical bone under 2D orthogonal cutting conditions. The aim of the research was to develop a predictive model of the sawing forces and to report them as a function of depth of cut. To obtain the micro geometric input, a heterogeneous anisotropic model was created from several images taken via an optical microscope of the cortex of adult mid-diaphysal bovine femur. In order to identify the various regions representing the micro-architecture of cortical bone, such as osteons, Haversian canals, lamellae and lacunae, MATLAB was utilized for intelligent image processing based on pulsed coupled neural networks. After each micro-phase in the image was assigned the proper mechanical properties, these material-tagged micro-features were imported into the finite element method (FEM) solver. Results from the simulation were correlated to cutting force data that was determined experimentally. Experiments were conducted with individual stainless steel saw blade teeth that were removed from a typical surgical saw blade. The teeth were 0.64 mm thick, with a rake and clearance angle of −10 and 60 degrees, respectively. Representative of clinical conditions for power bone sawing, depths of cut per tooth between 2.5 micrometer and 10 micrometer were investigated. The simulated cutting forces from the mFEM model compared favorably to the experimental data.Copyright

Distribution of Porosity in Cortical (Bovine) Bone

Pores (namely lacunae, clusters of canaliculi, Haversian canals, and resorption cavities) are present throughout cortical bone. This paper characterizes the area fraction (AF, %)) of each type of these pores as function of distance from the bone’s geometric center while noting the region in which such pores are located: midcortical or periosteal.Optical slides (at 20X) are taken from 2 cortical bone biopsies named bone 1 and bone 2 and cut at mid-diaphysis femur from 2 different (about 2 year-old) bovine cows. The slides are collected from posterior (pericortical) and anterior (intracortical) locations. The area of each of these biopsies is about 2.5mm × 3mm located near the outer cortex of the bone. In polar coordinates from the bone’s center, the areas cover radial distance of about 3.3 mm (of radius, R) and encompass an arc of 10°.Automated segmentation is used to locate and identify all pores in the optical slides the shapes of which are best fitted into ellipses. Values of area fraction, AF (%) of said fitted ellipses are then automatically calculated in secondary osteons for both regions. Variations in values of area fraction AF (%) are related to actual areas of pores (based on their defining equations).Observations suggest that area fractions (%) of all pores (but to lesser degree for Haversian canals), to significantly decrease linearly and in a steep fashion with R (statistically significant, p < 0.01) in the anterior region where osteonal growth is expected to have continued to develop. However, in the posterior region where osteonal growth appears to have matured, area fraction (%) values seem to have reached a steady state resulting in fairly flat behavior versus R. All observations are equally applicable for biopsies collected from bone 1 and bone 2.Copyright

Structural-Feature-Attribute-Based Segmentation of Optical Images of Bone Slices Using Optimized Pulse Coupled Neural Networks (PCNN)

In this work, we apply an image segmentation technique that uses pulse coupled neural networks to automatically discern the micro-features of cortical bone histology. In order to properly identify them, we exploit the geometric attributes of these micro features namely shape (i.e., circular or elliptical). These micro-constituent attributes are used as targets for the fitness function of the optimization method (particle swarm optimization, PSO) that was combined with PCNN along with an adaptive threshold, (T) that finds the best value for T between two segmented regions. The result is an optimal set of PCNN parameters that was found in this work to yield good-quality segmented pulses of the various micro-features of 2 different cortical bone images.Copyright

Segregation of Cortical Bone’s Haversian Systems via Automated Image Segmentation

The lamellar or Haversian system is comprised mainly of fundamental units “osteons”. Haversian canals run through the center of the osteons where one or more blood vessels are located. The bone matrix is comprised of concentric lamellae surrounding Haversian canals. Those lamellae are punctuated by holes called lacunae, which are connected to each other through the canaliculi supplying nutrients. Haversian canals, lacunae and canaliculi of the Haversian system constitute the main porosities in cortical bone, thus it is advantageous to segregate those systems in segmented images that will help medical image analysis in accounting for porosities.To the authors’ best knowledge, no work has been published on segregating Haversian systems with its 3 predominant components (Haversian canals, lacunae, and canaliculi) via automated image segmentation of optical microscope images. This paper aims to detect individual osteonal Haversian system via optical microscope image segmentation. Automation is assured via artificial intelligence; specifically neural networks are used to procure an automated image segmentation methodology.Biopsies are taken from cortical bone cut at mid-diaphysis femur from bovine cows (which age is about 2 year-old). Specimens followed a pathological procedure (fixation, decalcification, and staining using H&E staining treatment) in order to get slides ready for optical imaging. Optical images at 20X magnification are captured using SC30 digital microscope camera of BX-41M LED optical Olympus microscope. In order to get the targeted segmented images, utilized was an image segmentation methodology developed previously by the authors. This methodology named “PCNN-PSO-AT” combines pulse coupled neural networks to particle swarm optimization and adaptive thresholding, yielding segmented images quality. Segmentation is occurred based on a geometrical attribute namely orientation used as the fitness function for the PSO. The fitness function is built in such way to maximize the identified number of features (which are the 3 components of the osteonal system) having same orientation.The segmentation methodology is applied on several test images. Results were compared to manually segmented images using suitable quality metrics widely used for image segmentation evaluation namely precision rate, sensitivity, specificity, accuracy and dice.The main goal of segmentation algorithms is to capture as accurate as possible structures of interest, herein Haversian (osteonal) system. High quality segmented images obtained as well as high values of quality metrics (approaching unity) prove the robustness of the segmentation methodology in reaching high fidelity segments of the Haversian system.Copyright

Toward quantifying geometric microstructural differences between primary and secondary osteons via segmentation

Osteo-histological studies reveal that bones remodel themselves by removing mature bone tissue (bone resorption) from the skeleton and by forming new bone tissue (ossification). In cortical bone, remodeling results in secondary systems replacing of bone that has existed previously (primary bone). In histology slides, secondary bone appears as osteons with central Haversian canals while primary bone appears as interstitial lamellae occupying spaces between osteons. Cement lines refer to the boundaries demarcating the osteons. Although primary and secondary osteons differ micro-structurally, disambiguating such differences in histological studies presents a challenge. This paper aims to quantify such differences using automated segmentation utilizing artificial intelligence and geometric attributes: e.g., area (size), and compactness (shape). Preliminary findings suggest that vascular channels within primary osteons tend to be far more numerous but of smaller sizes than in secondary osteons.

Statistical and Physical Micro-feature-Based Segmentation of Cortical Bone Images Using Artificial Intelligence

At the micro scale, dense cortical bone is structurally comprised mainly of Osteon units that contain Haversian canals, lacunae, and concentric lamellae solid matrix. Osteons are separated from each other by cement lines. These microfeatures of cortical bone are typically captured in digital histological images. In this work, we aim to automatically segment these features utilizing optimized pulse coupled neural networks (PCNN). These networks are artificially intelligent (AI) tools that can model neural activity and produce a series of binary pulses (images) representing the segmentations of an image. Two segmentation methods were used: one statistical and another based on the physical attributes of the microfeatures. The first, statistical-based segmentation method, cost functions based on entropy (probability of gray values) considerations are calculated. For the physical-based segmentation method, cost functions based on geometrical attributes associated with microfeatures such as relative size, shape (i.e., circular or elliptical) are used as targets for the fitness function of network optimization. Both of these methods were found to result in good quality segregation of the microfeatures of micro-images of bovine cortical bone.

Application of Homogenization Theory to Study the Mechanics of Cortical Bone

Homogenization theory is utilized to study the effect on the axial stiffness of secondary osteons in cortical bone due to the presence of micro porous features (e.g., lacunae, canaliculi clusters, and Haversian canals). Specifically, 2 geometric characteristics were used to describe these features within the secondary osteons: volume fraction (% porosity) and shape (circular- or elliptical-shaped). Such information was determined for each individual porous feature from an image segmentation methodology developed earlier by Hage and Hamade. For each feature, aspect ratio vectors (or arrays of ratios for each individual porous feature) were used to classify each pore inhomogeneity as cylindrical, elliptical or irregular shape. Two prominent homogenization theories were used: the Mori-Tanaka (MT) and the generalized self-consistent method (GSCM). Using the results of image segmentation, it was possible to calculate the respective Eshelby tensors of each porous feature. To calculate the isotropic stiffness tensors for matrix (Cm) and pores (Cp) the Young’s modulus and Poisson’s ratio for the matrix (Em, νm) were assigned as obtained from literature and as those of blood (Ep=10MPa, νp= 0.3), respectively. The effective elastic stiffness tensors (C*) for the secondary osteons were obtained from which axial Young’s modulus was obtained as function of volume fraction (% porosity) of each pore type and their individual shapes. The normalized axial Young’s modulus was found to 1) significantly decrease with increasing volume fraction (%) of porosity and 2) for the same % porosity, to slightly decrease (increase) with increasing ratio of circular-shaped to elliptical-shaped (elliptical-shaped to circular-shaped) porous features. These findings were validated using experimental micro-indentation study performed on secondary osteons.Copyright