Last Updated on April 26, 2026
Evidence-based content
All content on ThyForLife is developed from peer-reviewed scientific literature, published clinical guidelines, and data from accredited medical research institutions.
Thyroid nodules are quite common. They are small lumps in the thyroid gland of some people. High-resolution ultrasound can detect nodules in up to half of adults, many of whom have no symptoms at all. Fortunately, most nodules are benign, but a small percentage can be malignant (cancerous). In previous blog articles, we talked about thyroid nodules and the main types of thyroid cancer. In this article, however, we turn our focus toward an emerging approach that aims to address the challenge clinicians face in deciding which nodules need further testing (such as fine-needle aspiration biopsy) and which can be safely monitored.
Artificial intelligence (AI) has become increasingly popular over the past couple of years, as many people have become more familiar with it, especially through the use of generative AI. However, there is so much more to AI than just generative AI. Machine learning (ML), which is a subfield of AI, is beginning to revolutionize scientific research as well as healthcare. This includes thyroid care.
In this article, we will answer the following questions:
How does the thyroid ultrasound work?
Ultrasound is the first-line imaging tool for evaluating thyroid nodules because it is non-invasive, radiation-free, widely available, and relatively inexpensive. During a thyroid ultrasound, clinicians assess features such as echogenicity (how bright or dark the nodule appears), margins (how smooth or irregular the nodule is), shape, presence of calcifications, and vascular patterns (the patterns of blood vessels).
To standardize interpretation, systems like TIRADS (Thyroid Imaging Reporting and Data System) were developed. While helpful, ultrasound interpretation still depends heavily on the experience of the operator. Inter-observer variability, therefore, remains a real problem, even among experts.
Your ultrasound results belong in one place
Store your ultrasound results alongside labs and symptoms for the full picture. Plus self-care tools for sleep, mental health, and physical wellness designed specifically for people with thyroid conditions. Free on iOS and Android.
Download ThyForLife for FreeWhat does AI actually do in thyroid ultrasound?
In medical imaging, AI typically refers to deep learning (DL). Just as ML is a subfield of AI, so is DL a subset of ML. DL uses neural networks trained on thousands of labeled images.
For thyroid ultrasound, AI systems are trained to detect nodules automatically, extract quantitative features (that are invisible to the human eye), classify nodules (as low- or high-risk for malignancy), and support clinical decision-making. These AI systems are able to classify nodules by means of a risk-scoring process. The AI model is able to convert the extracted features into a risk category using an established scale (for example, the ACR TI-RADS descriptors that radiologists use). Once trained, these models can analyze new ultrasound images in seconds.
Is there enough evidence that AI actually improves accuracy?
One major benefit of AI-assisted thyroid ultrasounds is reducing unnecessary biopsies. Fine-needle aspiration is safe, but it is still invasive and anxiety-inducing for some patients. AI-assisted ultrasound interpretation can help identify truly low-risk nodules more confidently, reduce “just-in-case” biopsies, and focus healthcare resources where they are most needed. This could realistically be especially valuable in high-volume endocrine clinics and resource-limited settings in the world.
But, what does the current scientific literature say about the accuracy of AI-assisted ultrasounds?
Overall, recent scientific reviews and studies show promising performance. A lot of these publications find that computer-aided diagnosis (CAD) and AI tools often reach diagnostic accuracy comparable to experienced radiologists for many nodules, and can reduce variability between readers. For example, according to a review article (published in the International Journal of Endocrinology in 2022) that reviewed about 25 studies with over 17,000 thyroid ultrasound images, the AI-assisted diagnostic tools were very effective, and especially so for under-50-year-olds. However, performance generally varies across studies and datasets.
Also, several AI-based software tools for TI-RADS ultrasound analysis have received FDA 510(k) clearance. This is an important step because it means regulators found evidence the software is safe and effective for its intended use. These tools are designed to automate the classification of thyroid nodules and reduce inter-observer variability. Some examples include See-Mode Technologies (received FDA clearance in 2024), Koios DS Thyroid (received FDA clearance in 2021), and Medo-Thyroid (received FDA clearance in 2021).
Are there any limitations or safety considerations worth considering?
It’s worth pointing out that many AI models perform best on the kinds of images and patient populations they were trained on. When images come from different machines, operators, or ethnic groups that weren’t well represented in training, accuracy can drop. That’s why prospective, multi-center validation matters. This simply means the AI is tested on new ultrasound scans from multiple hospitals to confirm it works reliably in real-world clinical settings before widespread clinical adoption.
This also goes to show that these systems may not be perfect for every case, as AI struggles with unusual presentations (such as very inflamed glands, complex cysts, or rare tumor types). Human review remains essential. Also, lower-quality images, such as from older machines or inexperienced operators, reduce accuracy. Think “Garbage in, garbage out.”
Will AI replace clinicians?
The simple answer is no. This is because thyroid nodule management is more than just an image-classification problem. It also depends on factors such as patient age and history, thyroid function tests, and symptoms and risk factors, among others. AI can definitely support consistency and accuracy, but the need for clinical judgment remains immutable. A practical analogy is likening AI to an assistant. It simply cannot replace the doctor’s judgment.
Key takeaways
- Here’s a practical scenario of how AI can help with thyroid ultrasounds. Imagine you have a small (8 mm) thyroid nodule caught on ultrasound:
- The ultrasound technologist captures standard images.
- The AI flags the nodule and suggests TI-RADS 2 (low risk) based on shape, smooth margins, and lack of calcifications.
- The radiologist reviews the images and the AI’s annotations, agrees, and recommends routine ultrasound follow-up rather than immediate biopsy
- As a result, you feel less anxious, as this is one less invasive procedure to worry about, and there is a clear plan for monitoring.
- The ultrasound technologist captures standard images.
- Despite its promise, AI is not a magic solution. It is still a tool that must be used together with clinical judgement and established guidelines.
- As research advances and validation continues, AI-assisted thyroid imaging is likely to become a routine part of modern thyroid care.
Related Articles



