Automated detection and delineation of mitochondria in electron micrographs of human skeletal muscles

Morphometric measurements of mitochondria in human skeletal muscles provide useful information relating to tissue oxidative energy production, nutrition, exercise, and aging. Morphometric data such as area, perimeter, long axis, and short axis can be obtained by delineating individual mitochondria in electron micrographs. However, manual counting and delineating of individual mitochondria is a formidable task. The purpose of this study was to develop a fully automated computer algorithm for quantifying mitochondrial morphometry in electron micrographs. The algorithm locates mitochondria with a two-dimensional matched filter and then traces the borders of individual mitochondria. The delineation is accomplished by edge detection along radial lines launched outwards from the center of each mitochondrion. Shape descriptors applied to delineated mitochondria are used to reject likely false-positive selections. The results show that the fully automated algorithm detects mitochondria with a false-positive rate of 2% and a false-negative rate of 36%. The errors are easily and rapidly corrected by user intervention using a second semiautomated delineation algorithm. Morphometric measurements collected with the automated algorithm are equivalent to those obtained manually by human experts. The algorithm significantly improves the speed of image analysis and it also provides copious quantities of high-quality mitochondrial morphometric data.