Development of a smart portable ultrasound system for real-time guidance of minimally invasive procedures

Abstract

Needle visibility is crucial in the diagnosis and treatment of non-communicable diseases. This work focuses on developing a smart portable ultrasound system that uses machine learning techniques to aid needle visibility during minimally invasive procedures and therefore make the diagnosis of noncommunicable diseases by radiologists more precise and safe. Segmentation and localisation models were developed and trained, validated and tested on a stratified dataset of 1410, 525,172 images respectively. The localization model obtained an average inference time of 0.027s and a mAP@0.5 of 0.988. On the same system, the segmentation model produced an IoU of 0.7273 with an average inference time of 0.018s. Clinical validation of this system will be the focus of future work. This study presents an advancement in the development of a smart system for the real-time localization and segmentation of needles under ultrasound guidance.