Texture analysis of post breast cancer lymphedema ultrasound images obtained using a portable device - A pilot study

Background: Lymphedema is a common complaint of post-breast cancer treatment. Some prior ultrasound imaging studies of lymphedema-affected upper extremities focused upon estimating the thickness of the dermis and subcutis areas. In contrast, however, the reliability and validity of texture features derived from ultrasound images obtained using a low-cost portable ultrasound device have not been reported. This study examined the reliability and validity of the first and second order quantitative image texture measures, average pixel intensity and entropy, and compared their site-specific correlations. Methods and results: A total of 10 ultrasound images of an ipsilateral affected elbow and the corresponding unaffected contralateral elbow were obtained using a portable ultrasound device. A Graphical User Interface software package for image analysis was developed and tested. Entropy and average pixel intensity at six adjacent regions of interest corresponding to the dermis and subcutis areas were calculated. Entropy had smaller coefficients of variation for each image compared to average pixel intensity with the intracluster correlation reliability coefficient for entropy being twice that of average pixel intensity. Both measures discriminated between the affected and unaffected sites (p<0.001). However, the correlation between the two measures differed for affected (r=-0.060, p=0.667) and unaffected (r=+0.595, p<0.001) sites. Conclusions: The second order measure entropy is preferable to that of the first order average pixel intensity as a measure of image texture to characterize ultrasound images of extremity lymphedema.