A lung disease diagnosis algorithm based on 2D spectral features of ultrasound RF signals.

Lung diseases are commonly diagnosed based on clinical pathological indications criteria and radiological imaging tools (e.g., X-rays and CT). During a pandemic like COVID-19, the use of ultrasound imaging devices has broadened for emergency examinations by taking their unique advantages such as portability, real-time detection, easy operation and no radiation. This provides a rapid, safe, and cost-effective imaging modality for screening lung diseases. However, the current pulmonary ultrasound diagnosis mainly relies on the subjective assessments of sonographers, which has high requirements for the operator's professional ability and clinical experience. In this study, we proposed an objective and quantifiable algorithm for the diagnosis of lung diseases that utilizes two-dimensional (2D) spectral features of ultrasound radiofrequency (RF) signals. The ultrasound data samples consisted of a set of RF signal frames, which were collected by professional sonographers. In each case, a region of interest of uniform size was delineated along the pleural line. The standard deviation curve of the 2D spatial spectrum was calculated and smoothed. A linear fit was applied to the high-frequency segment of the processed data curve, and the slope of the fitted line was defined as the frequency spectrum standard deviation slope (FSSDS). Based on the current data, the method exhibited a superior diagnostic sensitivity of 98% and an accuracy of 91% for the identification of lung diseases. The area under the curve obtained by the current method exceeded the results obtained that interpreted by professional sonographers, which indicated that the current method could provide strong support for the clinical ultrasound diagnosis of lung diseases.