Enhancing thyroid nodule malignancy prediction: integrating photoacoustic imaging-derived SO2 with clinical and ultrasound data.

Abstract

Photoacoustic imaging (PAI) has shown promise in diagnosing thyroid nodules. However, current methods rely on subjective visual assessments, lacking quantitative precision. This study evaluates the diagnostic accuracy of PAI in distinguishing benign from malignant thyroid nodules. The study integrates PAI with ultrasound and clinical data to improve prediction accuracy. A total of 407 thyroid nodules were analyzed, divided into training and testing sets (8:2). Dual-wavelength PAI was used to measure the oxygen saturation (SO2) values of lesions. Predictive factors were identified through logistic regression, resulting in three models: Mod-1 (clinical factors), Mod-2 (clinical + ultrasound factors), and Mod-3 (clinical + ultrasound + PAI-derived SO2 factors). Diagnostic performance was assessed using the area under the curve (AUC) and the DeLong test. Malignant lesions exhibited significantly lower oxygen saturation values (77.25 vs. 65.08, P < .01). The AUC for average oxygen saturation parameter was 0.829. In the testing cohort, the AUCs for Mod-1, Mod-2, and Mod-3 were 0.696, 0.947, and 0.974, respectively, with Mod-3 outperforming the others. PAI-derived SO2 provides a quantitative, noninvasive approach for thyroid nodule diagnosis. Combining PAI with clinical and ultrasound data enhances malignancy prediction, aiding personalized management.