An Autotuning Diaphragm Assessment System Based on Ultrasound Image Tracking

Ultrasound imaging provides a noninvasive and convenient way of visualizing diaphragmatic morphology and movement for patients during ventilator use. The high professional requirements, time-consuming setup for transducer location and ultrasound beam orientation and manual image selection and analysis may affect the reliability of the diaphragm functional assessment. In this study, an easy-to-use ultrasound system for diaphragm assessment was proposed based on an autotuning probe and an automated image-tracking model. The NeuroEvolution of Augmenting Topologies was used to train the diaphragm positioning algorithm. A state-of-the-art video tracking algorithm was fine-tuned with our ultrasound dataset, and fully automated through guidance from a trained multi-task U-net. Post-processing quantified the diaphragm contour to obtain features, such as the mean thickness and thickening rate, for potential diaphragm functional assessment. Results on two subjects demonstrated the accuracy and feasibility of our system for automated diaphragm quantification and potential assessment in ICU.