Validation of Freehand 3D Tomographic Ultrasound and TI-RADS-Generating Machine Learning Algorithm to Evaluate Thyroid Nodules

The following thesis aimed to evaluate whether 3-dimensional tomographic ultrasound and machine learning algorithm, as integrated into the Infinitytm software, are able to increase interobserver agreement for measurements and classifications of thyroid nodules in the clinical setting. Software generated lobe measurements and TI-RADS classifications were compared to those made with conventional ultrasound using two physician raters, and ratings were compared between physicians. Interrater agreement (Cohen’s Kappa) was found to be low for TIRADS scores for conventional 2D scans (κ = .23, p = .01) and was found to be even lower (κ = .15, p = .12) for automated scans. Use of the Infinity software was found to have poor intramodality agreement for both physicians 1 (κ = -0.07, p = .28) and 2 (κ = .19, p < .01). Many differences were observed between dimensional measurements that should have been the same between raters and scan types. Most notably, measurements of length and volume were found to agree poorly with conventional scans due to artifacts from poor tracking. Despite offering advantages in flexibility and visualization, the Infinitytm system was deemed not ready for clinical use due to gross inaccuracies, particularly in length and volume measurements.

Embargo status: Restricted until 06/2027. To request the author grant access, click on the PDF link to the left.