396551 Wavelet-Based De-Noising of Cellular Images

Monday, November 17, 2014
Galleria Exhibit Hall (Hilton Atlanta)
Roshan S. Patel, Yiider Tseng and Stephen Arce, Department of Chemical Engineering, University of Florida, Gainesville, FL

Wavelet analysis is a prevalent tool for image compression (JPEG2000 uses wavelets to compress images), data storage and de-noising, yet its application in biophysics has been largely unexplored. One challenge in the measurement and analysis of image based parameters for biological specimens is the removal of undesired noise from the image, which creates ambiguity and inaccuracy when segmenting via thresholding. The accuracy of measurements of these image based parameters could be improved significantly by incorporating wavelets as a de-noising tool.

After image acquisition, a discrete wavelet decomposition (with a variety of wavelet families) was used to separate the underlying frequencies in the cell image, thus isolating the high frequencies, characteristic of noise. The high frequencies were then attenuated and the image is reconstructed to provide a de-noised image. This method was used to de-noise fixed phase contrast and fixed fluorescent images of cells and nuclei. This alone was able to improve segmentation, allowing more accurate measurement of imaged-based parameters. In combination with Spatial Filtering, the segmentation was even greater improved and was comparable to methods currently used in image processing.

The results from wavelet de-noising can be benchmarked against existing de-noising methods such as Gaussian blur and anisotropic diffusion to determine what has optimal performance in terms of information retrieval (in particular, image based parameters) from cell images. Some basic cell information includes cell parameters such as the cell area, perimeter, boundary curvature, length-to-width ratio or average protrusion length. With better resolution, these cell parameters can reveal important, and more accurate, details about cell migration, drug resistance and proliferation that are otherwise unavailable due to image noise. This improved cell tracking and segmentation can be a powerful tool to advance medical studies in topics such as cancer development, guided nerve regeneration and wound healing.


Extended Abstract: File Not Uploaded
See more of this Session: Undergraduate Student Poster Session: Education & General Papers
See more of this Group/Topical: Student Poster Sessions