Radar degli Studi Clinici

Part A: Prospective Thyroid Imaging Study Objective: Visualization and Image Quality Assessment of Key Neck Structures Using Ultrasound Visibility Scale and Radiologist Comparison Form

1 Visibility Scale (Primary Component) Independent, blinded radiologists will rate the visibility of each structure using a 3-point scale: 0 = Not Visible (structure cannot be identified) 1. = Partially Visible (structure identifiable but with limited definition) 2. = Fully Visible (structure clearly identified with complete anatomic definition) Scores from each reviewer will be averaged for analysis. 2. Radiologist Comparative Image Quality Form (Secondary of Primary Outcome) Radiologists will complete a structured comparison form evaluating: Size (agreement between devices; measured in mm) Margins (sharp, indistinct, irregular) Edge Definition (well-defined vs. poorly defined) Overall Visibility (device-to-device comparison: better, worse, equivalent) These assessments will be recorded for each thyroid nodule and relevant neck structure. The primary endpoint is the difference in mean visibility scores and qualitative image quality ratings between the investigational and SOC ultrasound

This will take place at month 3 to end of month 7 of study.

Part B: Machine Learning Model Development Objective Develop and internally evaluate a machine-learning model that identifies and segments thyroid nodules and predefined neck structures (carotid, jugular, vagus nerve, trachea, thyroid nodule) on image

Outcome Measure 1: Segmentation Accuracy Performance of the machine-learning model in segmenting thyroid nodules and neck structures (carotid artery, internal jugular vein, vagus nerve, trachea). Accuracy will be evaluated using overlap-based metrics, including the Dice Similarity Coefficient and Intersection-over-Union (IoU). These unitless measures quantify agreement between the model-generated segmentation and expert-annotated reference masks. Outcome Measure 2: Visibility and Detection Performance Performance of the model in identifying the presence or absence of thyroid nodules and key neck structures on ultrasound images. Detection quality will be assessed using standard classification metrics, including sensitivity, specificity, positive predictive value, negative predictive value, and area under the ROC curve (AUC). These metrics quantify diagnostic accuracy. Outcome Measure 3: 3-D Reconstruction and Focus-Tracking Feasibility Feasibility of generating consistent 3-D thyroid

At month 4 until then end of the study, month 12.