Search Results

Sonio Suspect is intended to assist interpreting physicians, during or after fetal ultrasound examinations, by automatically identifying and characterizing abnormal fetal ultrasound findings on detected views, using machine learning techniques.

The device is intended for use as a concurrent reading aid on acquired images, during and/or after fetal ultrasound examinations.

The device provides information on abnormal findings that may be useful in rendering potential diagnosis.

Patient management decisions should not be made solely on the results of the Sonio Suspect analysis.

Sonio Suspect is a Software as a Service (SaaS) solution that aims at helping interpreting physicians (designated as healthcare professionals i.e. HCP in the following) to identify abnormal fetal ultrasound findings during and/or after fetal ultrasound examinations.

Sonio Suspect is a web application accessible from any device connected to the internet. It can be accessed on a tablet, computer or any other support capable of providing access to a web application.

Sonio Suspect can be used by HCPs as a concurrent reading aid on acquired images, to assist them during and/or after fetal ultrasound examinations of gestational age (GA): from 11 weeks to 41 weeks. A concurrent read by the users means a read in which the device output is available during and/or after the fetal ultrasound examination.

The way Sonio Suspect is built allows the HCP to use it at any moment. The software can process any Ultrasound image file uploaded by the HCP, at any time.

Sonio Suspect can be connected through API to external devices (as an ultrasound machine) to receive images.

Sonio Suspect workflow goes through the following steps:

As soon as an image is automatically received, it is automatically detected and associated with a view (and can be manually re-associated by the HCP). Then abnormal fetal ultrasound findings linked to the view are evaluated and displayed, individually, with one of the following status:

• Suspected (abnormal findings identified on the image);
• . Not Suspected (abnormal findings not identified on the image);
• . Can't be analyzed (abnormal findings not evaluated due to one or several structures not detected or if the fetal position selected is "other or unknown" while it's required to evaluate the abnormal finding).

Each abnormal finding status can be manually overridden to Present or Not Present by the user.

Here's a summary of the acceptance criteria and the study proving the device meets them, based on the provided text:

Acceptance Criteria and Reported Device Performance

Detailed Study Information:

1. Sample size used for the test set and the data provenance:

   • Standalone Test Set: 8,745 fetal ultrasound images from 1,115 exams.
   • Clinical Test Set: 750 fetal ultrasound images (between 11 and 41 weeks) evaluated by each reader, from 287 distinct exams.
   • Data Provenance: The standalone test set included data from 75 sites, with 64 located in the United States. The clinical test set included data from 47 sites, with 37 located in the United States. This indicates a mix of US and OUS (Outside US) data, explicitly representing the intended use population. The study was retrospective.

2. Number of experts used to establish the ground truth for the test set and the qualifications of those experts:

   • The document implies that ground truth for the clinical study was based on expert consensus, as it refers to a "fully-crossed multiple case (MRMC) retrospective reader study" where readers provide a "binary determination of the presence or absence of an abnormal finding." However, the exact number of experts explicitly establishing the ground truth for the test set (as opposed to participating as readers) or their specific qualifications for ground truth establishment are not explicitly stated in the provided text. The readers themselves were:
     • 13 readers: 5 MFM (Maternal-Fetal Medicine), 6 OB/GYN (Obstetrician-Gynecologists), and 2 Diagnostic radiologists.
     • Experience: 1-30+ years' experience.

3. Adjudication method for the test set:

   • The document states that in the clinical study, "For each image, each reader was required to provide a binary determination of the presence or absence of an abnormal finding and to provide a score representing their confidence in their annotation." It also mentions "two independent reading sessions separated by a washout period." While this describes the reader process, it does not explicitly describe an adjudication method (like 2+1 or 3+1) used to establish a definitive ground truth from multiple expert opinions. It implies that the ground truth was pre-established for the images used in the reader study.

4. If a multi reader multi case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance:

   • Yes, an MRMC comparative effectiveness study was done.
   • Effect Size: The study demonstrated a significant improvement in reader accuracy. The Area Under the Curve (AUC) for readers:
     • Without AI assistance ("Unassisted"): 68.9%
     • With AI assistance ("Assisted"): 90.0%
     • This represents a significant difference (effect size) of 21.9% in AUC.

5. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:

   • Yes, a standalone performance testing was conducted.
   • The results are detailed in Table 3, showing an average sensitivity of 93.2% and specificity of 90.8% for abnormal finding detection.

6. The type of ground truth used:

   • Implicitly, expert consensus or pre-established clinical diagnosis. For the standalone study, the robust sensitivity and specificity metrics suggest comparison against a definitive "ground truth" for the presence or absence of abnormal findings. For the clinical study, readers compared their findings against this ground truth. The document does not specify if pathology or outcomes data were directly used to define the ground truth for every case, but it's common for such studies to rely on a panel of experts or established clinical reports to define the ground truth for imaging-based diagnoses.

7. The sample size for the training set:

   • The sample size for the training set is not explicitly stated. The document mentions that the global validation dataset for standalone testing "was independent of the data used during model development (training/internal validation) and the establishment of device operating points," implying a separate training set existed, but its size is not provided.

8. How the ground truth for the training set was established:

   • This information is not explicitly provided. It can be inferred that similar methods to the test set (e.g., expert review and consensus) would have been used, but no specifics are given in the text.

Ask a specific question about this device