# APBA Kısa Özet
AI-assisted sonography for thyroid nodule diagnosis shows promising diagnostic performance with area under the curve values ranging from 0.78 to 0.97. Integration with TI-RADS and reported time savings highlight potential clinical utility, though current evidence is limited by study design and risk of bias.
# Çalışma neyi araştırdı?
This systematic review synthesized evidence from 30 studies published between January 2018 and June 2025 to evaluate the diagnostic accuracy, clinical utility, and workflow integration of AI-assisted sonography for thyroid nodule assessment. The primary objective was to determine the tool's potential for improving diagnostic consistency, reducing unnecessary biopsies, and assisting clinical decision-making. Eligible studies compared AI-based models using histopathology or fine-needle aspiration as reference standards.
# Yöntem
Studies were identified through database searches following PRISMA 2020 criteria. Data extraction captured AI architecture, sonography mode or technique, study design, clinical impact measures (fine-needle aspiration reduction, interpretation time savings), and diagnostic metrics (sensitivity, specificity, area under curve). A total of 30 studies were included. Convolutional neural networks constituted the most common AI architecture. Diagnostic performance metrics were extracted, and risk-of-bias was assessed using standardized tools. Clinical utility was evaluated through reductions in fine-needle aspiration procedures and time savings per case interpretation.
# Temel bulgular
- Diagnostic accuracy: Area under curve values ranged from 0.78 to 0.97. Specificity ranged from 70.4% to 90.7%, and sensitivity from 75.6% to 94.0%.
- Clinical utility: AI integration with TI-RADS was associated with a 10% to 45% reduction in the number of fine-needle aspirations performed. Some studies reported between 2.5 to 3.1 minutes of interpretation time saved per case.
- Study characteristics: Convolutional neural networks were the most frequently used architecture. Study designs were predominantly retrospective, contributing to moderate-to-high risk of bias across the evidence base.
# Bulgular ne anlama geliyor?
The review indicates that AI-assisted sonography holds potential for improving thyroid nodule characterization and streamlining clinical workflows. The observed reductions in fine-needle aspirations and interpretation time suggest practical benefits for radiologists and clinicians. However, the moderate-to-high risk of bias in most included studies limits the strength of these conclusions. The findings call for cautious interpretation and highlight the necessity of robust prospective validation before widespread clinical adoption.
# Klinik önem
While AI-assisted sonography demonstrates diagnostic performance metrics that could support clinical decision-making, the current evidence base does not yet support practice-changing recommendations. The observed time savings and reduced biopsy rates are encouraging but require validation in prospective, multicenter trials. Clinicians should view these tools as complementary to, rather than replacements for, established diagnostic frameworks such as TI-RADS. The review underscores the importance of standardized reporting and explainable AI features to facilitate safe integration into routine thyroid imaging workflows.
# Sınırlılıklar
- Included studies exhibited heterogeneous AI architectures, sonography modes, and study designs, limiting direct comparisons.
- Risk-of-bias assessment showed 53.3% of studies to be low risk, 33.3% to be moderate risk, and 13.3% to be high risk, primarily due to retrospective design or small sample size.
- No prospective multicenter trials were included in the current evidence base.
- Standardized reporting and external validation were inconsistently applied across studies.
- Citation metrics are bibliometric attention signals and do not by themselves establish clinical validity or evidence quality.