See-Mode Augmented Reporting Tool, Thyroid (SMART-T)

The See-Mode Augmented Reporting Tool, Thyroid (SMART-T) is a machine learning-based software intended to enhance the analysis of thyroid ultrasound images by automatically localizing nodules and characterizing them using ACR TI-RADS descriptors. Clinical performance data indicates that SMART-T improves radiologists' accuracy in nodule localization, lexicon descriptor identification, and TI-RADS level agreement, thereby aiming to support more consistent and accurate thyroid ultrasound reporting.

See-Mode Augmented Reporting Tool, Thyroid (SMART-T) is a stand-alone, web-based software designed to assist medical professionals in analyzing thyroid ultrasound images for adult patients. It employs machine learning algorithms to process DICOM-compliant ultrasound images, automatically identifying and localizing suspicious soft tissue nodules. The software then characterizes these nodules by generating lexicon-based descriptors based on the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS)—including composition, echogenicity, shape, margin, and echogenic foci—along with a calculated TI-RADS level. SMART-T facilitates structured reporting by generating an initial report from its analysis, which clinicians subsequently review, modify, and approve, and can also be used for reading measurements and annotations in further ultrasound studies. The device's efficacy was evaluated through standalone tests and a multi-reader multi-case (MRMC) study involving 18 radiologists and 600 patient cases. The MRMC study demonstrated that radiologists using SMART-T achieved improved performance: nodule localization accuracy increased (e.g., 95.6% aided vs. 93.6% unaided), and characterization of all ACR TI-RADS lexicon descriptors was significantly enhanced (e.g., margin accuracy rose from 57.3% unaided to 73.5% aided). Agreement with ground truth TI-RADS levels also improved (e.g., 60.0% aided vs. 51.1% unaided overall), particularly for clinically significant higher TR levels. The standalone performance of the software itself was robust, showing high accuracy in tasks like localization (95.1%) and TI-RADS level agreement (63.8%), with consistent results across diverse patient and reader subgroups.

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 machine learning model's performance on specific TI-RADS descriptors, particularly 'Margin' (aided 73.5%, standalone 58.4%) and 'Echogenicity' (aided 77.4%, standalone 68.2%), where standalone performance is notably different from aided, impact the overall reliability and clinical utility of the generated TI-RADS level before radiologist review?; What were the specific criteria and processes used by the two expert US-board certified radiologists and the adjudicator to establish the 'ground truth' for localization and TI-RADS characterization, and how was inter-observer variability among these experts managed and quantified in the study?; The LROC analysis showed an AULROC improvement with the device (aided 0.758 vs. unaided 0.736 for IOU>0.5); how does this quantitative improvement translate to tangible clinical benefits, such as a reduction in the rate of unnecessary Fine Needle Aspirations (FNAs) or an increase in the detection of clinically significant malignancies?; Given that the dataset was 74% from the US and 81.5% female, what is the anticipated performance generalizability and potential variability when the SMART-T software is deployed in global populations with significantly different demographic characteristics, endemic goiter rates, or thyroid nodule prevalence?; While the study demonstrated improved TI-RADS level agreement and descriptor accuracy, what is the software's quantifiable impact on the radiologists' reporting workflow, specifically regarding the time taken to complete a thyroid ultrasound report and the consistency of reporting across radiologists with varying levels of experience?