SAT-342 Interim Analysis of AIBx Software for Thyroid Cancer Diagnosis from Ultrasound Images. A Multicenter International Study

Abstract Disclosure: R.D. Paparodis: None. A. Rizoulis: None. P. Anagnostis: None. J.C. Jaume: None. D.P. Askitis: None. D. Zianni: None. I. Androulakis: None. N. Angelopoulos: None. S. Livadas: None. A. Boniakos: None. D. Bantouna: None. Y.S. Guerra: None. K. Bayrakdar: None. G. Simeakis: None. A. Melhim: None. J. Thomas: None. A. Melhim: None. Introduction: AIBx is an artificial intelligence (AI) software designed to assist in diagnosing thyroid cancer by analyzing ultrasound images of thyroid nodules. Initial studies demonstrated high sensitivity. This study aims to validate AIBx' s diagnostic accuracy across various ultrasound devices that were not in the training data and identify factors influencing its predictive performance. Methods: Two ultrasound images (sagittal and transverse) of thyroid nodules were collected from patients undergoing fine-needle aspiration (FNA) or surgical resection at nine endocrine clinics in Greece and two in the United States. Corresponding FNA cytology (Bethesda categories 2, 5, or 6) and/or surgical pathology data were obtained. AIBx software analyzed the images and classified nodules as benign (BEN) or malignant (MAL). The software's outputs were compared with cytology and/or pathology findings. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Results: A total of 672 consecutive patients were evaluated, with 495 cases (449 BEN and 46 MAL) included in the analysis. Mean age was 53.5 and n=381 (77.0%) were females. Among the analyzed images (sagittal: n=490; transverse: n=488), AIBx demonstrated the following predictive metrics: overall sensitivity of 63%, specificity of 54.3%, PPV of 25.9%, and NPV of 96.8%. Concordant diagnoses between sagittal and transverse images were observed in 361 cases (72.9%), with sensitivity increasing to 78.4% and specificity to 74.6%. For sagittal images alone, sensitivity was 71.7%, specificity was 69.8%, PPV was 19.8%, and NPV was 96%. Transverse images alone yielded a sensitivity of 73.9%, specificity of 66.1%, PPV of 18.5%, and NPV of 96.1%. Conclusions: AIBx represents an innovative AI-based approach to distinguish benign from malignant thyroid nodules using a simple interface. This interim analysis of a multicenter, international study indicates moderate predictive performance for concordant cases with high NPV comparable to other studies. However, further optimization is required before AIBx can be effectively integrated into routine clinical decision-making. Presentation: Saturday, July 12, 2025