Lung ultrasound (LUS) is used to assess the state of lung surface, but is strongly operator-dependent, leading to reduced reproducibility of LUS analysis. Indeed, clinicians can guarantee only limited precision in intercepting optimal imaging plane (O-ImP). Hence, in this study, we experimentally assessed the possibility to automatically intercept O-ImP in LUS exams (i.e., ImP perpendicular to pleural plane) by extracting three features, utilized to guide a robotic arm (UR5e) handling an ultrasound probe. We conducted experiments by designing a highly controllable environment, imaged with linear probe connected to ULA-OP platform and held by UR5e, programmed to explore 328 positions having rotational angles (RA) from −20º to 20º (positions with RA = 0º correspond to O-ImP). Normalized log-scale B-Mode images were formed and a rectangular region of interest (ROI) was considered to compute mean intensity (MeIn) at each depth of ROI. MeIn was utilized to extract: 1. max of cross-correlation of MeIn with envelope of transmitted pulse; 2. variance of MeIn; 3. spatial −12-dB-bandwidth of MeIn. These three features were fed to genetic algorithms to solve optimization problems to guide UR5e toward O-ImP. Results showed strong potential of investigated features in automatically intercepting the O-ImP in LUS examinations, with average errors below 1º.
Mento et al. (Wed,) studied this question.