Dual-Channel Near-Infrared Autofluorescence Imaging and AI Analysis to Locate Parathyroid Glands (PTFinder)

Overview

This study tests PTFinder, a dual-camera system that makes parathyroid glands glow on screen by capturing their natural near-infrared autofluorescence. After a thyroid or parathyroid operation, the removed tissue is scanned with PTFinder and then checked again under normal white light. We will measure how fast (seconds) and how accurately the device finds real glands, confirmed by frozen pathology or a rapid PTH strip. About 180 adult patients at three Chinese hospitals will join. The imaging adds only a few minutes and does not change any part of the surgery. We will also record blood calcium and PTH at 1 h, 24 h, and 7 d after surgery to see whether better gland recovery lowers low-calcium risk.Collected images will also be used to train and test a deep-learning model for fully automated parathyroid recognition; model performance metrics constitute secondary outcomes.

Prospective, paired, observational study at three endocrine-surgery centers. For each specimen: PTFinder scan - stopwatch runs until the first gland is seen; all suspected glands are marked. White-light exam - second investigator, blinded to device results, inspects the same specimen. All marked tissue plus a random "negative" sample go to frozen section or rapid PTH strip; discrepancies are settled by permanent histology. Outcomes Primary - time to first gland; detection rate per specimen. Secondary - sensitivity, specificity, PPV, NPV; post-operative PTH and calcium at 1 h, 24 h, 7 d; incidence of hypocalcaemia. Sample size: 60 specimens per site (total = 180) gives 80 % power to detect a rise in detection rate from 70 % to 90 % (α = 0.05, McNemar). Analysis: Paired tests for efficiency, mixed-effects models for accuracy and lab values; multiple imputation for missing data. Safety \& Oversight: PTFinder emits low-intensity NIR light only on ex-vivo tissue; no patient exposure. Device is investigational, not under an FDA IDE. Ethics approval obtained from Fujian Medical University Union Hospital IRB (FJMU-IRB-2025-PTF); other sites use the same central IRB agreement. Data are stored in a secure eCRF. Collected images will also be used to train and test a deep-learning model for fully automated parathyroid recognition; model performance metrics constitute secondary outcomes.