Premium
Enhancing spot detection and reducing noise from digitized electrophoretic gel images using area processing filters
Author(s) -
Patton Wayne F.,
Tempst Paul
Publication year - 1993
Publication title -
electrophoresis
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.666
H-Index - 158
eISSN - 1522-2683
pISSN - 0173-0835
DOI - 10.1002/elps.11501401101
Subject(s) - pixel , filter (signal processing) , median filter , artificial intelligence , image processing , noise (video) , gaussian filter , blob detection , computer science , gaussian , computer vision , mathematics , optics , pattern recognition (psychology) , physics , image (mathematics) , edge detection , quantum mechanics
Twenty area processing filters and filter combinations were evaluated in an effort to optimize presentation of two‐dimensional electrophoretic profiles to the Laplacian spot finder for maximal spot detection sensitivity. Images of electrophoresis gels were obtained by digitizing polyacrylamide gels at 1024 × 1024 picture elements (pixels) resolution with 256 grey scale levels using the charge coupled device (CCD) camera of the Millipore Bio Image 110S computerized imaging system. The images were imported into an Apple Macintosh microcomputer and selectively enhanced by applying various area processing filters. Previously described least squares fit, low‐pass, Gaussian and median filters were used to reduce noise in the digitized images. These filters differ in that during the summation process the least squares template weighs the immediately adjacent pixels more heavily than the Gaussian template. The low‐pass filters weigh all neighboring pixels equally. Median filters replace the pixel of interest with the middle (median) value of the pixel neighborhood. An analysis of convolution filter sizes indicated that a 7 × 7 matrix was optimal for 22 cm × 22 cm gels. When using the median area processing procedure, however, the 3 × 3 filter was found to be superior to the 7 × 7 filter. The 7 × 7 least squares filter significantly improved detection of low abundance polypeptides while having only minimal effects on the high abundance polypeptides. The 7 × 7 Gaussian and 3 × 3 median filters also improved detection of low abundance polypeptides but reduced the integrated areas of the high abundance polypeptides and thus their integrated optical densities as well. The 7 × 7 least squares and the 3 × 3 median filters effectively increased spot detection, and reduced noise with minimal alteration in the modal distribution of spot areas and integrated optical densities in the image. They also preserved image quality (minimal blurring). The practical consequence of using the 7 × 7 least squares, 3 × 3 median or 7 × 7 Gaussian filters is an 8‐fold increase in sensitivity due to noise reduction, resulting in the detection of 60–65% more polypeptides in a typical silver stained gel.