Sonio Detect

Sonio Detect is an AI-powered, cloud-based software designed to enhance fetal ultrasound examinations by automatically detecting views, anatomical structures, and verifying image quality criteria in real-time. Its methodology combines machine learning, biometric analysis, and colorimetry to provide a concurrent reading aid for healthcare professionals, aiming to improve the completeness and standardization of obstetric ultrasound assessments.

Sonio Detect is a Software as a Service (SaaS) solution designed to assist healthcare professionals, including sonographers, OB/GYNs, Maternal-Fetal Medicine specialists, and fetal surgeons, in performing routine fetal ultrasound examinations in real-time across all trimesters (gestational age 11 to 37 weeks). Its primary function is to analyze fetal ultrasound images and clips using machine learning techniques to automatically detect standard obstetric views, identify anatomical structures within these views, and verify predefined quality criteria. The system operates by receiving DICOM instances from compatible ultrasound machines (GE, Canon, Samsung) via a locally installed Edge Software, which uploads the data to Sonio's cloud platform for processing; Sonio Detect's algorithm leverages artificial intelligence, biometric analysis from images, and colorimetry for 3D and Doppler images to provide concurrent reading support, aiming to ensure examination completeness and adherence to imaging protocols. The software highlights any unverified views or quality criteria, such as the presence or absence of specific anatomical structures (e.g., cavum septum pellucidum, cerebellum, heart chambers, major vessels) or correct zoom levels, allowing users to review and override the system's automated assessments. Quality criteria checked include the presence of key brain structures like the cavum septum pellucidum and cerebellum, and heart components like ventricles, atria, and major vessels. Performance testing on a dataset of 17,885 fetal ultrasound images demonstrated high sensitivity and specificity across various detection tasks: view detection (e.g., T1 views sensitivity 0.942, T2/T3 views sensitivity 0.919), anatomical structure identification within brain (sensitivity 0.857, specificity 0.963) and heart views (sensitivity 0.900, specificity 0.982), and verification of zoom levels for brain views (sensitivity 0.952, specificity 0.906). The system also showed high sensitivity in detecting 3D (0.980) and Doppler (0.963) images.

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 'Lecture of biometrics' and 'Colorimetry identification' component of the algorithm specifically improve performance beyond standard AI techniques for fetal ultrasound analysis?; What is the range of variability in performance when Sonio Detect is used with different compatible ultrasound machines (GE, Samsung, Canon) under real-world clinical conditions?; What is the learning curve for healthcare professionals to effectively use Sonio Detect, including the override functionality, and integrate it into their existing workflow?; How does the system handle ambiguous or poor-quality ultrasound images, and what is the threshold for when it cannot provide a confident detection or quality assessment?; Are there plans to expand the list of automatically detected views, anatomical structures, or quality criteria, particularly for more complex fetal anomalies or later gestational stages?