Optical diagnosis is the interpretation of images obtained during diagnostic procedures to give an instantaneous assessment of the nature of the tissue in view. In respiratory medicine, bronchoscopic procedures are the main diagnostic methods. In the last 2 decades, newer imaging modalities have been developed to improve the diagnostic yield of bronchoscopy. These include endobronchial ultrasound (EBUS), autofluorescence (AF) bronchoscopy and more recently narrow band imaging (NBI). Even with all these developments, histology of biopsies can be equivocal or negative despite
obtaining appropriate images of the suspect lesion(s).
Previous research in EBUS and NBI image analysis has been mainly descriptive and highly dependent on subjective expert bronchoscopist interpretation of images. AF has been more thoroughly investigated via the analysis of red to green ratios (R:G) of AF intensity. These bronchoscopic techniques lend themselves well to further objective image analysis with computer assisted software, thereby adding to the diagnostic information obtained at the time of the respective investigations.
The major hypothesis of this thesis is that objective image analysis of bronchoscopic techniques can provide further information in the clinical assessment of patients. In particular, such objective clinical information could be useful where a biopsy showed benign pathology, and the image analysis data was consistent with this, thereby allowing a conservative approach to be taken.
The overall aim of this thesis was therefore to establish optical diagnosis as an adjunct to histology to improve the yield of bronchoscopic investigations, thereby potentially reducing the number of procedures for patients.
(i) Greyscale texture analysis of EBUS images
Image J (developed at the U.S. National Institutes of Health and available on the Internet at http://rsb.info.nih.gov/nih-image/) and MATLAB (version 184.108.40.2067(R2009a), The Mathworks, Natwick, MA) were used to assess EBUS images based on first and second order greyscale texture features. For EBUS convex probe (EBUS-CP) images, the whole lymph node was assessed. For EBUS mini probe (mini probe) images, an expert defined region of interest methodology was used.
(ii) Clinical application of AF and NBI at panendoscopy.
AF and NBI bronchoscopy of the upper airways and bronchus was performed at panendoscopy to assess for sensitivity and specificity in detecting additional lesions to the known primary carcinoma. Impact on management was also assessed.
(iii) Colour profile analysis of narrow band imaging
Using an expert defined region of interest methodology, NBI images of the head and neck and bronchus were analysed for the following colour profiles: red, green and blue, hue saturation and brightness and the L*a*b* colour profiles.
(i) Greyscale texture analysis of EBUS-CP images found 5 features that differentiated between malignant and benign lymph nodes. The validation set correctly classified 44/51 (86.3%) lymph nodes. For EBUS-MP images, three features differentiated malignant and benign peripheral lesions. The validation set correctly classified 63/82 (76.8%) peripheral lung lesions.
(ii) Both AF and NBI were more sensitive than white light in the head and neck and bronchus regions for detecting additional lesions to the primary carcinoma. NBI was more specific than AF in both regions (p=0.003). 12/73 (16.4%) of patients had additional findings that directly impacted on immediate management.
(iii) Forty-three lesions of moderate dysplasia or worse in the head and neck were available for colour profile analysis. In the H&N, RGB colour profile analysis showed significant differences between ROI of biopsied lesions and ROI of presumed normal background mucosa with area under the receiver operator curves of > 0.80.
For the first time in bronchoscopic procedures, greyscale texture analysis has been effectively applied to EBUS images to assist in differentiating between benign and malignant aetiologies. Further larger studies would be warranted to demonstrate consistent results across different EBUS processors and physicians of varying experience.
NBI yielded significantly better specificity than AF in a clinical setting based on expert opinion of abnormal lesions. Both modalities contributed to a change in management in head and neck cancer patients. Initial colour profile image analysis data is promising in this series of patients and demonstrates significant objective differences. Further numbers and a validation set would be beneficial to strengthen these initial findings.