Microscopic image analysis for quantitative characterization of muscle fiber type composition

Abstract

Skeletal muscles consist of muscle fibers that are responsible for contracting and generating force. Skeletal muscle fibers are categorized into distinct subtypes based on several characteristics such as contraction time, force production and resistance to fatigue. The composition of distinct muscle fibers in terms of their number and cross-sectional areas is characterized by a histological examination. However, manual delineation of individual muscle fibers from digitized muscle histology tissue sections is extremely time-consuming. In this study, we propose an automated image analysis system for quantitative characterization of muscle fiber type composition. The proposed system operates on digitized histological muscle tissue slides and consists of the following steps: segmentation of muscle fibers, registration of successive slides with distinct stains, and classification of muscle fibers into distinct subtypes. The performance of the proposed approach was tested on a dataset consisting of 25 image pairs of successive muscle histological cross-sections with different ATPase stain. Experimental results demonstrate a promising overall segmentation and classification accuracy of 89.1% in identifying muscle fibers of distinct subtypes.

Introduction

Muscular disorders are usually associated with the loss of muscle strength that is due to the loss of muscle mass and a reduced number and size of individual muscle fibers. Muscle mass is usually moderately correlated with muscle function, and other factors, such as physiological conditions and structural factors, which are also playing important roles in assessing muscle performance. In our continuous efforts to identify factors relevant to the muscle function and structure, we focus on performing automated characterizations of the number and the size of skeletal muscle fiber subtypes using the adenosine triphosphatase (ATPase) stained histological muscle cross-sections. Our goal is to characterize the composition of muscle fiber subtypes in terms of their numbers and sizes, to profile the overall muscle architecture and to correlate the structural information with the results from functional characterizations. The identified correlations between structural and functional parameters of muscle will significantly enhance our understanding of the factors affecting muscle quality.

Skeletal muscle consists of individual muscle fibers that are responsible for contraction and generating force. As summarized in Table 1, based on their characteristics such as contraction time, force production and resistance to fatigue, skeletal muscle fibers are classified as slow twitch, e.g., type I, and fast twitch, e.g., types IIA, IIB, and IIX muscle fibers [1]. The quality and the strength of a muscle are strongly related to its fiber type composition, which can be revealed by a histological evaluation, i.e., placing a sectioned tissue sample under a microscope and performing a visual examination to identify relative number and size of distinct muscle fiber subtypes.

Present histochemical techniques make it possible to differentiate distinct fiber types based on the differences in the myofibrillar actomyosin adenosine triphospatase (ATPase) activity [2]. In clinical and laboratory practice, ATPase stain is applied to successive histology sections at different pH levels and the observations from successive cross-sections are incorporated in order to classify muscle fibers into distinct types. Fig. 1 shows images of two successive serial cross-sections of a 26 month old rat's gastrocnemius muscle stained with ATPase stain at (a) pH = 4.5 and (b) pH = 10.4. Fig. 1(a) demonstrates type I (dark intensity), type IIa (light intensity) and type IIb-x (medium intensity) and Fig. 1(b) demonstrates type I (light intensity), type IIa (dark intensity) and type IIb-x (medium intensity) muscle fibers as summarized in Table 1.

Characterizing the muscle fiber type composition requires identification of individual muscle fiber in terms of their number and size (i.e., the cross-sectional area (CSA)) among distinct muscle fiber subtypes. This requires a knowledgeable operator to visually analyze the successive histological cross-sections with different ATPase activity (e.g., pH = 4.5 and pH = 10.4). Manual assessment is extremely time-consuming since it requires delineation of hundreds of fibers by hand and identifying the correspondences between individual muscle fibers between successive slides with different ATPase activity. This study aims at developing a quantitative image analysis framework to characterize the muscle fiber type composition from digitized skeletal muscle tissue samples.

Recent developments in high throughput whole-slide scanners allow digitization of entire tissue samples at high magnifications up to 40×. As a result, we can now incorporate automated image analysis systems in evaluation of tissue morphology to aid clinicians in their decision making mechanism. Providing computational tools to measurable biologically relevant features, histopathological image analysis may help extracting more objective and more accurate quantitative information regarding the relevant tissue morphology. In case of muscle histology evaluation, we develop an automated evaluation tool that allows not only to evaluate the entire tissue area on each slide, but also to evaluate a very higher number of tissue samples in a reasonable time period that would not be feasible otherwise.

Tissue slides used in our study are scanned at 5× magnification and the average resolution of a resulting digitized image is around 11,000 × 9000 with an approximate disk size of 280 Mb. Considering the memory limitations in processing these large images, for computational issues, we simply divide the whole-slide image at smaller image tiles and process each tile independently. Therefore, the developed image analysis flowchart explained in Section 2 considers the processing of a single image tile manually cropped from a whole-slide image.

There is substantial amount of published literature on microscopy image analysis for a wide variety of modalities ranging from confocal and electron microscopy to optical bright field microscopy [3]. Applications on stained cytology and histopathology specimens are relatively more recent and have been accelerated after the recent developments on high-throughput digital whole-slide scanners [4], [5], [6]. A common step to analyze stained histopathology samples at microscopic scales is the segmentation of the tissue image into basic cytological components. Feature space clustering algorithms, where each cytological component is represented by one or a mixture of multivariate distributions in a feature space usually constructed using the color and the texture information are by far the most widely used methods [7], [8], [9], [10]. Typically, this is followed by a feature construction step where clinically and statistically relevant measurements are obtained in order to achieve further higher-level tasks such as classification, detection, recognition, and grading [10], [11], [12], [13], [14], [15]. In general, due to highly textured appearance of tissue structures in histopathology and cytology imagery, image gradient information usually does not provide additional information to improve the segmentation process. In muscle histology images, on the other hand, prominent boundaries between individual muscle fibers bear discriminative information. By exploiting the gradient information between the muscle fibers and the connective tissue structures, Brox et al. and Kim et al. proposed edge-based active contour approaches for the segmentation of individual muscle fibers from digitized histology images [16], [17]. However, although they presented a segmentation accuracy of 98% in their application, H&E-stained histological sections can only be used to identify the overall muscle fiber size distribution and do not provide these measures among distinct subtypes, which is the goal of our study. In a similar study using active contour models, Klemencic et al. proposed a semi-automated segmentation approach from ATPase stained muscle histology images [18]. However, their system requires significant manual interaction (i.e., the reader is required to manually determine the center location of each muscle fiber); hence they may not be practical to analyze a large number of high-resolution microscopic images. Dryden et al. proposed a Markov chain monte carlo algorithm that represents the muscle fibers as polygons and models the segmentation as the maximization of the posterior probability based on the tessellation of the spatial domain [19]. In a more recent study, Karen et al. developed a software for muscle fiber type classification and analysis, however again, their software requires significant manual interactions both for the registration and the segmentation steps [20]. There are also similar applications in the domain of electron microscopy image analysis. Jurrus et al. proposed an image analysis framework to segment and track axons from transmission electron microscopy images using an active contour based approach that heavily relies on manual inputs regarding the center locations of each axon to be segmented [21]. Cuisenaire et al. used a simple thresholding followed by the construction of a zonal graph that is used to apply a set of semantic rules that only allows identification of isolated axon candidates [22]. Several studies have also been conducted on the registration of muscle images and will be discussed later in Section 2.3 [23], [24]. To the best of our knowledge, this is the first study to develop a comprehensive, automated system to characterize muscle fiber composition from histological images. In our preliminary work, we have developed and presented an automated registration system [25]. This study is extending our earlier work to a comprehensive system for automated evaluation of muscle histology slides.

The rest of the paper is organized as follows: Section 2 explains the steps of the proposed image analysis pipeline in detail, including segmentation, registration and classification. Section 3 presents the experimental results and discussions. Finally, Section 4 provides a summary and conclusions.