ARTIFICIAL INTELLIGENCE TO REDUCE FALSE POSITIVES AND FALSE NEGATIVES IN THE DIAGNOSIS OF THYROID CANCER

Abstract

Thyroid cancer is the most common endocrine neoplasia and presents diagnostic challenges related to the high rate of false positives and false negatives, leading to unnecessary interventions or delays in treatment. This work reviews the application of artificial intelligence (AI) in the diagnosis of this disease, addressing methods that optimize the analysis of ultrasound and cytology images. Machine learning algorithms, convolutional neural networks (CNNs) and tools such as AIBx and AI-TIRADS have demonstrated greater accuracy, specificity and the ability to reduce biopsies and invasive procedures. Studies have compared the effectiveness of these systems in relation to traditional methods, such as TIRADS and clinical assessments, highlighting the potential of AI in stratifying risks and identifying benign and malignant nodules with greater accuracy. Models such as MRF-Net have achieved high sensitivity and specificity, while semi-automated systems, such as S-Detect, have improved performance in contexts with less specialized professionals.
Despite the advances, limitations include the need for greater sensitivity in cases that require surgical decisions and the adaptation of models to specific populations. It is concluded that AI is a promising tool in the clinical management of thyroid cancer, but requires additional validations for its safe and effective application on a large scale.