Detection of Cervical Intraepithelial Neoplasia (CIN) by Spectral Similarity Mapping Histology and Computer-Assisted Classification
AON Joseph, G Galliano, S Bose, DL Farkas. Cedars Sinai Medical Center, LA, CA; Cedars Sinai Medical Center, LA, CA
Background: Screening for cervical carcinoma and its precursors by standard cervical pathology has known problems. There is a need for new methods for better detection and discrimination of CIN. Using new optics and/or automated image analysis to reduce subjectivity should provide effective screening tools for early detection of cervical cancer. However, long-term studies using available cervical images are needed to validate the potential of new algorithms in discriminating CIN from normal tissue.
Design: This clinical study examined the diagnostic potential of spectral imaging and automated image processing in patients with CIN with the extreme grades of negative and CIN III. We investigated the use of spectral imaging - based on combining spectroscopy and digital imaging - for analysis of dyes and enzyme precipitates on pathological specimens. A pixel-by-pixel spectrum-based color classification of single and double color immunocytochemical staining of H&E in paraffin-embedded cervical tissue samples was performed. Ten stage CIN III, and 5 negative patients were sampled for a total of 53 image sets. Images were collected from 450-800 nm for data cubes consisting of 46 wavelengths. All slides were imaged by using a Zeiss microscope and our modified commercial spectral instrument, at 10X objective magnification, yielding up to 700 MB of uncompressed data per set. Established spectral unmixing techniques separated regions of interest and quantified presence of indicators. The cases were also examined and marked by two experienced pathologists.
Results: The results demonstrate that spectral imaging can reliably identify chromogenic dyes in a single bright-field microscopic specimen. Regions of interest (ROI) were segmented, quantified, and used to build a spectral reference database of disease. Separate microscopic fields from the same patient can be analyzed using this spectral reference library, to check reproducibility.
Conclusions: We conclude that spectral imaging is a reliable and robust method for pixel color recognition and classification in CIN. Our data further indicate that the use of spectral imaging a. can increase the number of parameters studied simultaneously in pathological diagnosis, b. may provide quantitative data (such as positive labeling indices) more accurately, c. may solve segmentation problems currently faced in automated screening of tissue specimens and d. can enhance the objectivity and reproducibility of histopathologic diagnosis.
Wednesday, March 11, 2009 1:00 PM
Poster Session VI # 227, Wednesday Afternoon