Radial Endobronchial Ultrasound Greyscale Texture Analysis Using Whole-Lesion Analysis Can Characterise Benign and Malignant Lesions without Region-of-Interest Selection Bias

Background: Radial-probe endobronchial ultrasound (RP-EBUS) is predominantly used clinically for the localisation of peripheral pulmonary lesions prior to biopsy. However, the RP-EBUS image itself contains information that can characterise the aetiology of lesions.

Objectives: The aim of this study was to show the utility of RP-EBUS image analysis using unconstrained regions of interest (ROIs) that utilise more image information and eliminate ROI selection bias.

Methods: We developed custom software to analyse RP-EBUS images digitally captured during clinical procedures. Unconstrained ROIs were mapped onto lesions. We computed first-order greyscale image statistics of minimum, maximum, mean, standard deviation and range of pixel intensities, and entropy. We also computed second-order greyscale texture features of contrast, correlation, energy and homogeneity. The results of image analysis were compared to gold-standard tissue diagnosis. Features from expert- and non-expert-defined ROIs were also compared.

Results: Eighty-five images were analysed (38 benign and 47 malignant). Five greyscale features were significantly different between benign and malignant lesions. Benign lesions had higher mean (p < 0.01) and maximal (p < 0.001) intensity, greater range (p < 0.001) of pixel intensities and greater entropy (p < 0.01). The highest positive predictive values were associated with maximal (87.8%) and range of pixel (83.8%) intensities. There were no significant differences between expert- and non-expert-defined ROIs.

Conclusion: RP-EBUS image analysis using unconstrained ROIs eliminates ROI selection bias and can characterise benign and malignant lesions with an accuracy of up to 85%.