Investigate the performance of artificial intelligence based clinical decision support system in the assessment of thyroid nodules compared to current practice and explore the levels of acceptability for patients and stakeholders.
A third of patients with thyroid nodules currently undergo an unnecessary needle biopsy because ultrasound scans cannot reliably tell benign from cancerous growths. This matters because the number of detected thyroid nodules has surged with increased imaging, yet current risk-stratification systems rely on subjective human interpretation of ultrasound features. Different operators often classify the same nodule differently, and even when a biopsy is performed, the cell sample sometimes yields no clear answer. The result is low specificity—many patients endure invasive procedures for ultimately harmless nodules. The researcher will test whether an artificial intelligence algorithm can outperform standard ultrasound classification by identifying patterns invisible to the human eye. In a prospective study, the AI’s risk scores will be compared against biopsy results and post-surgical tissue analysis. Focus groups with patients and clinicians will explore whether people trust an AI-driven recommendation and what information they need to accept it. If the AI proves accurate and acceptable, the NHS could reduce unnecessary biopsies, cut patient anxiety, and save resources. An expert panel will produce evidence-based recommendations for integrating AI into the thyroid nodule pathway.
View original technical description
Background The increasing use of medical imaging has resulted in a significant increase in the investigation of thyroid nodules. Current stratification models, based on ultrasound features, are used to determine the need to perform fine needle aspiration (FNA) and assess the risk of thyroid cancer. Inter-observer variability, between ultrasound operators, and the limited ability of cytology to reach a definitive answer, in some nodules, gives low specificity. The use of Artificial Intelligence (AI), as a decision aid, within medical imaging is being increasingly researched; the ability of AI to automatically identify complex patterns imperceptible to experienced operators is of particular interest. I propose a mixed method research design to evaluate feasibility, accuracy, acceptability and impact of using AI in comparison to the current patient pathway. Research Question Investigate the performance of artificial intelligence based clinical decision support system in the assessment of thyroid nodules compared to current practice and explore the levels of acceptability for patients and stakeholders. Aims and Objectives To evaluate existing literature and health economic evidence regarding the use of AI in the investigation of thyroid nodules, to inform clinical feasibility To perform a prospective study to validate an AI algorithm against a standard practice operator assigned British Thyroid Association (BTA) ultrasound classification and gold standard cytological diagnosis. To explore the barriers and facilitators influencing stakeholders and patients acceptance of AI through focus group interviews To consider and explore the key points and overall themes of the research project through a synthesis of findings with an expert panel. Methods and Timeline for delivery Work Package 1 (months 1-6) Systematic literature and health economic evaluation review To assess the feasibility of using AI risk stratification in clinical practice compared to current patient pathway. Work Package 2 (months 6-18) A quantitative assessment of the accuracy of AI compared to the current standard of practice BTA ultrasound classification, FNA cytology and post-surgical histology. Work Package 3 (months 8-24) Qualitative multi-centre focus group interviews of a range of healthcare professionals, stakeholders and patients to determine their perspective regarding the use of AI and the associated barriers, facilitators and attitudes. To gain an understanding of the preferred form and type of information patients would like to receive about using AI as a diagnostic tool in thyroid US. Work Package 4 (months 24-34) A purposively selected expert panel will participate in a modified Delphi consensus process to review sets of statements based on work package 1-3 and agree upon a ≥75% agreement on a 9-point Likert scale to generate evidence-based recommendations for the integration of AI-assisted thyroid nodule assessment into clinical practice Anticipated impact and dissemination Publications and dissemination events, considering equality, diversity and inclusion, will be organised at the end of each work package. I will work with PPI groups and charities to disseminate findings from work package 3 and develop wider themes for upscaling and implementation. The anticipated impact of the project is to generate evidence-based recommendations for the integration of AI-assisted thyroid nodule assessment into clinical practice.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know