Publication

Abstract : Novice sonography practitioners use a variety of training devices for the retention and enhancement of their sonographic-probe manipulation skills. Conventional training devices utilized by sonographic practitioners are devoid of force-sensing capability, despite the inevitable application of force, a critical parameter in sonography. Besides, none of the contemporary sonographic training systems are outfitted to classify the fundamental sonographer actions: sliding, sweeping, compression, rotation, rocking, and fanning. Action classification is an essential aspect of the initial training of a sonographer. Training devices endowed with force-sensing capabilities are pricey and complex haptic devices, unaffordable for individual users. The core research of this paper lies in the design and development of a novel, simple, force/torque (F/T) sensor-based, and artificial intelligence (AI)-enabled training device. The forces and moments on the F/T sensor during probe manipulations were mathematically modelled to delineate various sonographic actions. The proposed device processes real-time data from positional sensors and a custom-designed F/T sensor for sensor fusion, employing deep learning (DL) techniques. The experimental results show that the proffered device exhibited sensitivities of 6.1 mV/Nm for measuring moment around the Y-axis, 0.062 mV/N for force along the Z-axis, and 5.4 mV/Nm for moment around the X-axis. The resolutions of force and moment measurement were 0.098 N and 0.006 Nm, respectively, corroborating pragmatic measurement of force variations. The DL model deployed in the device attained 99 % accuracy, demonstrating the capability of accurate classification of sonographic actions. The device can serve as an affordable pre-training module and a worthy recommendation to original equipment manufacturers.