Automated image analysis of skeletal muscle fiber cross-sectional area

Abstract

Morphological characteristics of muscle fibers, such as fiber size, are critical factors that determine the health and function of the muscle. However, at this time, quantification of muscle fiber cross-sectional area is still a manual or, at best, a semiautomated process. This process is labor intensive, time consuming, and prone to errors, leading to high interobserver variability. We have developed and validated an automatic image segmentation algorithm and compared it directly with commercially available semiautomatic software currently considered state of the art. The proposed automatic segmentation algorithm was evaluated against a semiautomatic method with manual annotation using 35 randomly selected cross-sectional muscle histochemical images. The proposed algorithm begins with ridge detection to enhance the muscle fiber boundaries, followed by robust seed detection based on concave area identification to find initial seeds for muscle fibers. The final muscle fiber boundaries are automatically delineated using a gradient vector flow deformable model. Our automatic approach is accurate and represents a significant advancement in efficiency; quantification of fiber area in muscle cross sections was reduced from 25-40 min/image to 15 s/image, while accommodating common quantification obstacles including morphological variation (e.g., heterogeneity in fiber size and fibrosis) and technical artifacts (e.g., processing defects and poor staining quality). Automatic quantification of muscle fiber cross-sectional area using the proposed method is a powerful tool that will increase sensitivity, objectivity, and efficiency in measuring muscle adaptation.

Fig. 1.

Dystrophin detection to measure muscle fiber cross-sectional area (CSA). Representative image from dystrophin immunohistochemistry (IHC; red) of plantaris cross-sections that delineates the sarcolemma and DAPI staining (blue) to visualize nuclei. Scale bar = 100 μm.

Fig. 2.

Intermediate results which demonstrate the steps of semi-automatic analysis of CSA using Zeiss AxioVision software with manual corrections. A: initial segmentation using Axiovision software. B: high magnification of boxed region in A. C: segmented fibers after a second parameter correction during Axiovision segmentation, which includes identifying inter- and intra-myofibrillar spaces. D: high magnification of boxed region in C. E: segmented image following third parameter corrections requiring interactive processing. Scale bar = 100 μm.

Fig. 3.

Comparison between manual annotation and automatic segmentation algorithm. Final segmented image with semiautomatic procedure followed by manual annotation (A), high magnification of A (B); and automatic segmentation algorithm (C). CSAs are shown in squared micrometers with ID region representing each individual fiber.

Fig. 4.

Automatic image segmentation procedure. The procedure involves three steps: 1) ridge detection, 2) muscle fiber seed postprocessing, and 3) gradient vector flow deformable model.

Fig. 5.

Intermediate image segmentation results show the steps of automatic analysis of CSA with the proposed segmentation algorithm. A: original image. B: ridge detection results. C: detection of muscle fiber seeds. Seeds are represented with white color overlaid on the original image. D: final automatic muscle fiber segmentation results.

Fig. 6.

1st, 10th, 15th, and 50th (left to right) iterations of the deformable model evolution results overlaid on the original image.

Fig. 7.

Image segmentation following wheat germ agglutinin (WGA) staining. A: Texas Red conjugated WGA binds to proteoglycans in the extracellular matrix providing fiber boundaries for accurate CSA determination with our approach. Scale bar = 100 μm. B: final segmented and quantified image with the automatic procedure. C: high magnification of B.

Fig. 8.

Image segmentation following supra-physiological overload of the plantaris. A: dystrophin immunoreactivity shows disorganization and fiber size heterogeneity in plantaris muscles subjected to synergist ablation surgery. Scale bar = 100 μm. B: final segmented and quantified image with the semiautomatic procedure with manual correction. For this image, due to biological artifacts, more than 40 min was required for analysis. C: final segmented and quantified image using the automatic algorithm that required 15 s for the entire process.

Fig. 9.

Segmentation comparison of an image with poor staining quality. A: weak dystrophin immunoreactivity required extensive correction. Scale bar = 100 μm. B: final segmented and quantified image using the semiautomatic procedure followed by manual annotation. More than 40 min was required due to poor staining quality. C: final segmented and quantified image using the automatic algorithm that shows the identical number of fibers compared with manual annotations.