UltraSight AI Guidance

The UltraSight AI Guidance system is a software application designed to empower non-expert medical professionals to acquire standard cardiac ultrasound images by utilizing artificial intelligence. Its methodology involves real-time, on-screen visual cues for transducer manipulation and image quality assessment, based on machine learning algorithms trained to emulate sonographer expertise.

The UltraSight AI Guidance is a software accessory for compatible diagnostic ultrasound systems, designed to assist medical professionals without specialized sonography training in acquiring two-dimensional transthoracic echocardiography (2D-TTE) images from adult patients. It guides users through the acquisition of ten standard cardiac views (PLAX, PSAX-AV, PSAX-MV, PSAX-PM, AP4, AP5, AP2, AP3, SubC4, and SC-IVC). The system operates as a mobile application leveraging machine learning and artificial intelligence to provide dynamic, real-time instructions on transducer positioning and orientation. Key features include a "Quality Bar" that continuously scores image diagnosability based on ACEP guidelines, "View Detection" to identify if the heart is visible and guide navigation or signal to hold position, and "Probe Guidance" which offers on-screen graphical cues for transducer movements like rotation, tilt, rock, and slides, without explicit text instructions. Non-clinical performance testing validated the software according to IEC 62304 and demonstrated strong standalone performance for its AI algorithms: the Quality Bar achieved a mean AUC of 0.86 and PPV of 0.93 for diagnosability classification; View Detection showed a mean AUC of 0.988 for differentiating system states; and Probe Guidance cues achieved a mean AUC of 0.821. Human factors validation confirmed usability for intended users. Clinical performance was assessed in pilot and pivotal studies. The pivotal study involving 240 subjects showed that non-expert users, with UltraSight AI guidance, acquired images of sufficient quality for cardiologists to assess key cardiac parameters (LV size/function, RV size, pericardial effusion) in 93-100% of cases by majority agreement. Secondary endpoints also showed high rates of sufficient visual quality, affirming that non-expert users can acquire diagnostic-quality cardiac ultrasound clips.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the AI algorithm adapt to variations in patient anatomy or challenging acoustic windows not extensively covered in the training datasets?; What is the learning curve for non-expert users to achieve consistent diagnostic-quality image acquisition across all 10 cardiac views using the UltraSight AI Guidance?; Could the system's performance be further enhanced by integrating automated suggestions for adjusting ultrasound machine settings (e.g., depth, gain) in addition to transducer manipulation guidance?; What are the specific performance metrics (e.g., sensitivity, specificity) of the "View Detection" feature for differentiating between the 10 targeted cardiac views, especially in anatomically similar views?; How does the system's guidance accuracy and image quality output compare when used with different compatible ultrasound systems beyond the Philips Lumify S4-1 probe and Samsung Tab S6/S7 tablets?