Diagnostic efficacy of backscattering intensity measurements in optical coherence tomography of cervical intraepithelial dysplasia

• Published in: Lasers in surgery and medicine 2012-01, Vol.44 (1), p.11-19
• Main Authors: Gallwas, Julia, Mortensen, U., Gaschler, Rebecca, Stepp, H., Friese, K., Dannecker, C.
• Format: Article
• Language: English
• Subjects: Adult; Aged; Biopsy, Needle; Cervical Intraepithelial Neoplasia - diagnosis; Cervical Intraepithelial Neoplasia - pathology; Cervical Intraepithelial Neoplasia - surgery; CIN-classifications; epithelial brightness; Female; Humans; Hysterectomy - methods; Immunohistochemistry; intraepithelial cervical dysplasia; Middle Aged; Observer Variation; optical coherence tomography (OCT); Prospective Studies; Reproducibility of Results; Sampling Studies; statistical analysis of diagnostic judgments; Tomography, Optical Coherence - methods; Uterine Cervical Neoplasms - diagnosis; Uterine Cervical Neoplasms - pathology; Uterine Cervical Neoplasms - surgery
• ISSN: 0196-8092; 1096-9101
• DOI: 10.1002/lsm.21146

Background and Objectives The aim of this study was to determine the diagnostic efficacy of backscattering intensity measurements in optical coherence tomography in identifying different grades of cervical intraepithelial dysplasia. Study Design/Materials and Methods OCT images were taken from 153 unsuspicious and suspicious areas of 30 fresh conisation and hysterectomy specimens, evaluated by two blinded investigators using a six‐grade classification (normal, inflammation, CIN1, CIN2, CIN3, squamous carcinoma) and later compared to the corresponding histology. Differences between judgments based on either the histology or the OCT images were investigated employing Correspondence Analysis (CA). Further, we explored the extent as to which backscattering intensity profiles of OCT images contained the essential information required for a reliable and valid diagnosis, using Linear Discriminant Analysis (LDA). Results The CA of histology‐ and OCT‐based judgments suggests that the diagnostic process may be characterized in terms of two stochastically independent underlying (“latent”) variables, the first of them reflecting the definiteness with which CIN classes are identified, the second reflecting a bias towards diagnosing inflammation on the side of the OCT‐based judgments. This finding is supported by the results of LDAs, where histology and OCT categorizations differ in particular with respect to the positions of inflammation and CIN1. Possibly, a second canonical variable has to be assumed accounting for the evaluation of carcinoma. Conclusions The systematic differences between histology‐based and OCT‐based diagnoses suggest that the use of available information is influenced by perceptual and/or cognitive biases. Apart from this it seems that the profiles appear to provide a remarkably large amount of information determining the main course of the diagnostic process. Lasers Surg. Med. 44:11–19, 2012. © 2012 Wiley Periodicals, Inc.