Ongoing Development of An Imaging Informatics System for Muscle
1. (11/15/2013) Three new journal papers reporting the preliminary results of the automated muscle morphometric measurments were accepted.
2. (10/20/2013) One new journal paper was accepted into Journal of Applied Physiology (JAP) reporting fiber-type specific CSA and myonuclei counting.
3. (10/10/2013) The project started from 10/2011, after two years of preliminary study, we have processed over 6000 H&E stained muscle specimens, and over 20000 digitized fluoresence muscle specimens from over 10 different instituations in the United States. We will start to fulfill more requests from outside institutions as the project moves forward.
4. (03/01/2013) One conference paper in International Symposium on Biomedical Imaging (ISBI, 2013) is accepted.
5. (01/10/2013) Our first journal paper on automatic muscle image analysis was accepted to Jounal of Applied Physiology (JAP).
Note: Please contact Lin Yang if you have questions with regard to the following experimental results.
1. Our algorithm can calculate the fiber CSA containing hundreds of muscle fibers in less than 15 seconds, while the semi-automatic method (using off-the-shelf software followed by human expert post-processing) requires an average of 25-40 minutes even for an experienced technician. This represents a turnaround time that is 100-160 times faster than current procedure.
2. In one of our pilot studies, it took an experienced technician 2 months to determine ber CSA on 900 muscle cross-sections, where a mean of 100 bers per cross-section (90,000 muscle fibers in total) were analyzed. Our turnaround using the proposed algorithm is several hours.
The BICI2 imaging room has a newly purchased Olympus VS120 five slide loader whole slide scanning microscopic system, which has been specifically designed for fast whole slide scan with high image resolution. The careful integration of all the components creates a highly flexible system that enables the user to acquire digital slides quickly and effortlessly.
The slides can be scanned at 2x, 10x, 20x or 40x magnification with automated tissue and multiple regions of interest detection. Virtual-Z is equipped with this imaging equipment. It is a multiple z planes scanning utility with virtual focusing feature that makes the VS120 microscope and scanning system ideal for scanning thick samples where depth of focus is important.
The Olympus VS120 is also able to perform the fluorescence whole slide scan. All of the VS120 fluorescence components are designed to interact seamlessly, producing a fully automated, high speed multi-channel fluorescence scanning system with excellent flexibility with simple operations. The output image formats are standard TIFF, big TIFF or JPG. The Olympus VS120 imaging equipment is maintained, managed, and operated by Dr. Yang and BICI2 staff members.
|The animation to show the procedure using automatic biomedical image analysis and imaging informatics to perform content based image retrieval to search for myositis cases (DM, PM, and IBM) that exhibit similar image markers. The query image patch is taken from a patient with dermatomyositis (DM). All the retrieved cases are patients who exhibit similar contents measured by the automatically discoverd image markers.|
The preliminary segmentation results for H&E stained image. The left columns represent the original images and the right column denotes the automatic segmentation results.
The cell boundaries are highlighted using green line. The cross-section area of each cell is overlaid on the image. We also use different color to present the position of the myonuclei such as center (red) and inner boundary (yellow).
The automatic H&E segmentation results for challenging cases. The left columns represent the original images and the right column denotes the automatic segmentation results. Please note the artifacts and sample defects exhibited in the digital images during sample preparitions.
The cell boundaries are highlighted using blue line. The muscle on the boudaries are automatically rejected by the segmentation algorithm to avoid inaccurate CSA estimation.
Frozen muscles were sectioned (7 micron), air dried, and stored at -20C. To delineate fibers, fresh frozen sections were immunoreacted with the dystrophin antibody, followed by Texas Red-conjugated secondary antibody. Sections were post-fixed in 4% paraformaldehyde and then stained with DAPI to visualize nuclei. Three to five washes with phosphate buffered saline were performed between each step.
The preliminary segmentation results for fluoresence image. The left column represents the original image which contains over 900 muscle fibers. The right column denotes the automatic segmentation results. The cell boundaries are highlighted using green line. The cross-section area of each cell is overlaid on the image.
In order to delineate fibers, fresh frozen sections were immunoreacted with the dystrophin antibody, followed by Texas Red-conjugated secondary antibody. Sections were post-fixed in 4% paraformaldehyde and then stained with DAPI (10nM) (4',6-diamidino-2-phenylindole, Invitrogen, Carlsbad, CA) to visualize nuclei. Three to five washes with phosphate buffered saline were performed between each step.
For fiber type muscle cross sections, fresh frozen sections were immunoreacted with antibodies against myosin heavy chain isoforms I, IIa and IIb, followed by immunoglobulin specific secondary antibodies conjugated to different fluorophores. Three to five washes with phosphate buffered saline were performed between each step.
The automatic quantification by fiber typing is shown in the left figure. Both the cross-sectional area and fiber types are automaticlly calculated. The cell boundaries are highlighted with white lines. The final result is illustrated in (D).
|The graphic user interface prototype. The top window represents the image segmentation results with automatically delineated boundaries overlaid on the muscle fibers. The bottom window shows content based image rank retrieval results that identify muscle images exhibiting similar morphometric features to a query image.|
The content based image rank retrieval results using cross sectional area histogram (CSAH). Left: The query image. Right: the CBIR rank retrieval results. In the right panel, the first two rows represent the most similar cases (segmented image is on the top of CSAH distribution map) and the last two rows denote the most different muscle specimens measured with CSAH.
The retrieval is conducted on an image repository that contains 900 digitized muscle specimens with roughly 50,000 automatically segmented muscle fibers. It is obvious that the most similar cases exhibit comparable muscle fiber cross sectional area distributions.
Myosin type 2B staining for a wild type mouse. The cross-sectional area (CSA) by fiber typing are automatically calculated. One example out of 14 whole slide scanned cases are presented in the left. The top is the original image and the bottom is the automatic segmentation results. The automatic delineated cell boundaries are overlayed on the original image. The muscle fiber with red contour represents the positive and the fibers with green contour denotes the negative fiber. We also show Feret Minimal Diameter with blue lines.
Previously it will take a doctor's lab weeks or months for manual quantification of the whole slide scanned mouse specimens with tens of thousands of fibers in each case. The automatic algorithm can return all the results that contain CSA by fiber typing and Feret Minimal Diameter of each individual muscle fiber in several hours without human intervention involved.
The automated morphometric measurements of CSA for a dystrophin-deficient (mdx) mouse model. The upper panel is the original image. The bottom panel denotes the automated detection of fibers with central nuclei and measurement of fiber-type-specific CSA. The boundaries of 894 muscle fibers are highlighted using red and green lines. Fibers with central nuclei (627 in total) are automatically recognized and labeled with red contours, and the fibers without central nuclei (267 in total) are automatically recognized and marked with green contours. As one can tell, even for mdx mice which exhibit a large variation of fiber size and shape, we can still reliably provide the CSA and automatically detection of centralized nuclei fibers in 15 seconds with less than 2% errors.
|The software prototype of the CBIR and visualization functionalities. The left panel presents a testing image and three measures used for similarity comparisons. The right panel is the CBIR rank retrieval results that identify all the muscle images exhibiting similar measures as the query image sorted by the distances. The content based image rank retrieval results can be quickly sorted based on types, similarity, etc.|