CT-Free Quantitative Thyroid SPECT Based on Artificial Intelligence: A Prospective Multicenter Noninferiority Clinical Trial

Abstract

Background:

Hybrid single-photon emission computed tomography (SPECT)/computed tomography (CT) is used for the differential diagnosis of thyrotoxicosis, but its dependence on CT and complex analysis remains debated. This study evaluated whether CT-free thyroid SPECT using artificial intelligence (AI) could substitute for conventional SPECT/CT.

Methods:

This prospective multicenter noninferiority clinical trial (KCT0008387) included 152 patients with thyrotoxicosis from three tertiary referral hospitals. A pretrained AI model generated technetium thyroid uptake (TcTU) values using SPECT images without CT. These values were compared with TcTU derived from conventional SPECT/CT. The primary endpoint was diagnostic accuracy for Graves’ disease. Noninferiority was defined as a lower confidence interval (CI) limit greater than −10%. Secondary endpoints included diagnostic accuracy for destructive thyroiditis and prediction of antithyroid drug (ATD) prescription within one month.

Results:

Among 152 patients, 84 had Graves’ disease, 45 had indeterminate disease, and 23 had thyroiditis. For Graves’ disease, CT-free SPECT demonstrated 86.2% accuracy compared with 85.5% for conventional SPECT/CT. The lower bound of the CI of the accuracy difference was −3.0%, meeting the prespecified noninferiority margin. For thyroiditis and ATD prescription, the lower bounds were −7.4% and −3.2%, respectively. Omission of CT resulted in a 33.5% reduction of radiation exposure from 3.34 mSv to 2.22 mSv, and CT-free SPECT automatically generated TcTU values more quickly than SPECT/CT (40 minutes vs. 4 minutes).

Conclusions:

CT-free SPECT using AI demonstrated noninferior diagnostic performance for thyrotoxicosis compared with conventional hybrid imaging while reducing radiation exposure and analysis time.

Trial Registration:

https://cris.nih.go.kr (KCT0008387).