Feature analysis of ultrasound elastography image for quantitative assessment of cutaneous carcinoma

Abstract

Background

To evaluate the feasibility of using quantitative texture features computed from high frequency ultrasound and ultrasound elastography (USE) images in the discrimination of benign from malignant skin lesions.

Methods

A commercial ultrasound system with a 14 MHz transducer was used to visualize skin lesions requiring biopsy on clinical evaluation. Patients were enrolled over a 6-month period and imaged prospectively by operators blind to the histopathologic diagnosis. Anatomic ultrasound and USE imaging of the skin lesions was performed using a 2-4 mm gel standoff pad before biopsy and histopathologic evaluation. The ElastoAnalysis software developed for the texture analysis of USE images was provided by Hitachi. The software computes thirteen texture features within a region of interest (ROI), which have demonstrated promise in diagnostic characterization of liver fibrosis staging and in quantitative elastography of breast cancer. This approach has not yet been studied in the quantitative assessment of skin cancer. Results were retrospectively compared to the histopathologic diagnosis and a diagnostic criteria with the goal of maximizing sensitivity was evaluated for each textural feature.

Results

Of the 37 lesions included, among 30 patients who participated, 12 lesions were malignant and 25 were benign. Eleven out of thirteen textural metrics computed by the software were useful in differentiating benign from malignant lesions with 100% sensitivity and specificities ranging from 28% to 85%.

Conclusions

This feasibility study demonstrated that feature analysis of USE may be useful in quantitatively differentiating cancerous from benign primary solitary skin lesions prior to biopsy.