19 March 2013 Study and discrimination of human cervical tissue images through multifractal analysis
Author Affiliations +
Proceedings Volume 8577, Optical Biopsy XI; 85770W (2013) https://doi.org/10.1117/12.2005815
Event: SPIE BiOS, 2013, San Francisco, California, United States
Abstract
We report here a study of confocal microscope images to classify cervical precancers by a multifractal analysis. This study is performed using an inverted confocal microscope with laser scanning fluorescence imaging. The periodic structure of collagen present in the stromal region of cervical tissue gets disordered with progress in grade of dysplasia. This disorder is investigated through the β-exponent of a Discrete Fourier Transform (DFT) of the confocal images, enabling us to discriminate between the lowest and highest grades of dysplasia in human cervical tissue sections. The Holder exponent from 2D images further classifies various grades of dysplasia from normal tissue sections though Gd3 and Gd1 are indistinguishable. DFT however, clearly distinguishes Gd3 from Gd1. In addition to stromal images, epithelial images were also investigated for better classification. The cellular density of epithelium increases with depth for various grades of dysplasia and is not uniform. The Holder exponent, which measures multifractality, is higher for dysplastic tissue sections than for normal ones because of the above morphological differences. Extraction of subtle fluctuations from optical images through multifractal studies promise to be a powerful diagnostic technique.
© (2013) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Jaidip Jagtap, Pankaj Singh, Chayanika Pantola, Asha Agarwal, Kiran Pandey, Asima Pradhan, "Study and discrimination of human cervical tissue images through multifractal analysis", Proc. SPIE 8577, Optical Biopsy XI, 85770W (19 March 2013); doi: 10.1117/12.2005815; https://doi.org/10.1117/12.2005815
PROCEEDINGS
9 PAGES


SHARE
Back to Top