Decision-based AI Visual Navigation for Cardiac Ultrasounds

TL;DR

The study addresses noninvasive right atrial pressure assessment via ultrasound by enabling novices to locate the inferior vena cava (IVC) with AI-guided real-time localization. A decision-based video model detects IVC presence and a real-time localization algorithm annotates its position using internal feature maps, providing spatial guidance without explicit spatial training. Experimental results show robust performance: in a clinical trial with novice operators, the base navigation achieved $62\% \pm 9.98\%$ accuracy; localization reached $97\%$ on hospital data and $99.67\%$ zero-shot on the Butterfly iQ, plus $100\%$ localization when transferred to a RAP classification model. This work demonstrates real-time, annotation-enabled guidance, expanding ultrasound accessibility beyond hospitals and suggesting generalization to other vascular or anatomical targets, with deployment through the Butterfly iQ app.

Abstract

Ultrasound imaging of the heart (echocardiography) is widely used to diagnose cardiac diseases. However, obtaining an echocardiogram requires an expert sonographer and a high-quality ultrasound imaging device, which are generally only available in hospitals. Recently, AI-based navigation models and algorithms have been used to aid novice sonographers in acquiring the standardized cardiac views necessary to visualize potential disease pathologies. These navigation systems typically rely on directional guidance to predict the necessary rotation of the ultrasound probe. This paper demonstrates a novel AI navigation system that builds on a decision model for identifying the inferior vena cava (IVC) of the heart. The decision model is trained offline using cardiac ultrasound videos and employs binary classification to determine whether the IVC is present in a given ultrasound video. The underlying model integrates a novel localization algorithm that leverages the learned feature representations to annotate the spatial location of the IVC in real-time. Our model demonstrates strong localization performance on traditional high-quality hospital ultrasound videos, as well as impressive zero-shot performance on lower-quality ultrasound videos from a more affordable Butterfly iQ handheld ultrasound machine. This capability facilitates the expansion of ultrasound diagnostics beyond hospital settings. Currently, the guidance system is undergoing clinical trials and is available on the Butterfly iQ app.

Figures (4)

• Figure 1: Figure of the X3D and SlowFast video decision model architectures.
• Figure 2: One frame of a 1 million iteration SHAP run of an IVC video. The SHAP values can be displayed in the below level bar signifying the relative importance. Darker color represents a larger magnitude.
• Figure 3: Samples of the navigation algorithm run on real IVC videos. The top and bottom rows are from the high quality hospital and more affortable Butterfly iQ ultrasounds respectively. The green dot is the localized location of the IVC based on Algorithm 1.
• Figure 4: 3 examples of Algorithm 1 run on the RAP classification model. (a), (b), and (c) are example videos at low, medium and high RAPs respectively. The images are single frames of the video for the sake of display.