TIRADS Scoring for Thyroid Nodules: Accuracy and Limits

TIRADS, short for Thyroid Imaging Reporting and Data System, is a standardized framework that radiologists use to evaluate thyroid nodules found on ultrasound and decide whether they need a biopsy. The system assigns points based on how a nodule looks, then sorts it into a risk category ranging from benign to highly suspicious. Since the American College of Radiology published its version (ACR TI-RADS) in 2017, research has shown it can cut unnecessary biopsies of benign nodules by roughly 20 to 50 percent compared with other approaches, without missing a meaningful number of cancers.

How the Scoring Works

When a radiologist performs a thyroid ultrasound and spots a nodule, they evaluate five characteristics: the nodule’s composition (solid, cystic, or mixed), its echogenicity (how bright or dark it appears relative to surrounding tissue), its shape (whether it is taller than it is wide), its margins (smooth, irregular, or extending beyond the thyroid capsule), and the presence of echogenic foci (bright spots that can represent calcifications). Each feature earns zero to three points depending on how suspicious it looks. The points are added up, and the total determines the TI-RADS category.

The categories run from TR1 (benign, zero points) through TR5 (highly suspicious, seven or more points). TR1 and TR2 nodules are considered benign or not suspicious, respectively, and no biopsy is recommended. TR3 nodules are mildly suspicious. TR4 is moderately suspicious. TR5 is highly suspicious. For TR3 through TR5, whether a biopsy is actually recommended depends on the nodule’s size: a TR3 nodule typically needs to reach 2.5 centimeters before biopsy is suggested, while a TR5 nodule triggers a biopsy recommendation at just 1 centimeter. This size-gating is one of the key reasons ACR TI-RADS recommends fewer biopsies than competing systems.

How Well Does It Detect Cancer

A systematic review and meta-analysis of ACR TI-RADS found pooled sensitivity of about 89 percent and specificity of about 70 percent, with an overall diagnostic accuracy (area under the curve) of 0.86.¹ In plain terms, the system catches roughly nine out of ten cancers while correctly identifying about seven out of ten benign nodules as benign. The trade-off is deliberate: a small number of cancers are missed in exchange for sparing a large number of patients from invasive procedures on nodules that turn out to be harmless.

Looking at how individual categories perform, a large study found that TR3 nodules had a negative predictive value around 95 percent when compared with biopsy results, meaning that when the system said a TR3 nodule did not need a biopsy, it was right about 95 percent of the time.² On the other end, among TR5 nodules that went to surgery, about 71 percent were confirmed malignant. TR4 nodules occupied a middle ground: roughly 11 percent proved cancerous at surgery. These numbers illustrate why TR5 is taken seriously and TR3 usually warrants watchful waiting rather than an immediate needle.

Cutting Down on Unnecessary Biopsies

The strongest practical argument for TIRADS is that it prevents people from undergoing biopsies they do not need. Thyroid nodules are extraordinarily common. Ultrasound detects them in up to two-thirds of the adult population, yet only a small fraction harbor cancer. Before standardized risk systems existed, a lot of benign nodules were being biopsied simply because they were there.

One study comparing how radiologists managed nodules before and after adopting ACR TI-RADS found that the average number of nodules recommended for biopsy dropped from about 80 to about 57 per radiologist in the study cohort, a reduction of roughly 30 percent.³ Another analysis found that strict application of ACR TI-RADS would have eliminated more than half of all biopsies performed, with a false-negative rate of only about 2 percent.⁴ A separate single-institution study reported that over half of biopsies performed under previous local guidelines could have been safely deferred under ACR TI-RADS.⁵

This matters for patients beyond just the inconvenience of a needle biopsy. Thyroid biopsies occasionally produce indeterminate results, which can trigger further testing, repeat biopsies, or even diagnostic surgeries that turn out to have been unnecessary. Reducing the total volume of biopsies shrinks this cascade of downstream procedures.

Multiple TIRADS Systems Around the World

ACR TI-RADS is not the only game in town. Several countries and professional societies have developed their own versions, including the European Union’s EU-TIRADS, South Korea’s K-TIRADS, and China’s C-TIRADS. While they share the same basic goal of sorting thyroid nodules by cancer risk, they differ in how they get there. ACR TI-RADS uses its point-based scoring system across five ultrasound features, then adds size thresholds to determine whether biopsy is warranted. EU-TIRADS and K-TIRADS take a more pattern-based approach, assigning nodules directly into risk categories based on which suspicious features are present.⁶

In head-to-head comparisons, the systems tend to perform similarly for detecting cancer, though with different trade-offs. A multi-institutional comparison found that ACR TI-RADS recommends 25 to 50 percent fewer biopsies than the American Thyroid Association guidelines, EU-TIRADS, and K-TIRADS, largely because of its stricter size thresholds.⁷ ACR TI-RADS consistently shows higher specificity, meaning it is better at leaving benign nodules alone, though this comes at the cost of occasionally being slightly less sensitive. A multicenter prospective study of small nodules (under 1 centimeter) found that ACR TI-RADS, C-TIRADS, and K-TIRADS performed comparably, and all three outperformed EU-TIRADS in overall diagnostic accuracy for that size range.⁸

From a cost-effectiveness standpoint, at least one analysis found that ACR TI-RADS slightly edged out the American Thyroid Association’s 2015 guidelines, producing marginally better quality-adjusted outcomes at slightly lower cost.⁹ The difference was slim, though, and in practice the choice of which system a hospital uses often comes down to regional preference and what the local radiologists were trained on.

The Problem of Subjectivity

One persistent limitation of any TIRADS system is that different radiologists looking at the same nodule do not always agree on what they see. A study measuring how consistently radiologists scored the same nodules using ACR TI-RADS found that agreement on the final TI-RADS level was only about 50 percent, though agreement on the practical question of whether to recommend biopsy was higher, around 79 percent.¹⁰ Some features are easier to agree on than others: shape (taller than wide versus wider than tall) had the best agreement at about 87 percent, while echogenic foci had the worst at about 70 percent. Another study found only fair reliability for both raw TI-RADS scores and final categories among readers evaluating indeterminate nodules.¹¹

Training and structured consensus sessions help. One study showed that interobserver agreement improved significantly after a consensus reading session, regardless of which TIRADS system was used.¹² Still, the fact that two equally qualified radiologists can look at the same ultrasound image and arrive at different TI-RADS categories is something patients should be aware of, especially when a borderline score determines whether they are sent for a biopsy or told to come back in a year.

Artificial Intelligence and the Future of TIRADS

Given the subjectivity problem, researchers have been working to build AI systems that can score thyroid nodules more consistently. The results so far are promising but uneven. An early study using machine learning to create an “AI TI-RADS” found it achieved a slightly higher overall diagnostic accuracy than the standard ACR TI-RADS applied by expert readers (area under the curve of 0.93 versus 0.91), with meaningfully better specificity.¹³ The improvement was even more pronounced for less-experienced radiologists, suggesting AI could serve as a useful second opinion that brings everyone closer to expert-level performance.

More recent work has shown that AI-assisted TI-RADS classification can improve with software updates. One study found that after a system update, agreement between AI and human assessors rose from 73 percent to nearly 93 percent, and the number of cases where AI misinterpreted echogenic features (tiny bright spots that often trip up algorithms) dropped from about 20 percent to under 5 percent.¹⁴ The remaining disagreements tended to cluster around unusual presentations like autoimmune thyroiditis or hemorrhaged cysts, situations that challenge human readers too.

That said, AI is not ready to replace radiologists in this space. Current evidence suggests AI may have lower standalone accuracy than experienced radiologists, but it can meaningfully improve performance when used as an assist tool, particularly for identifying specific suspicious features that a busy reader might overlook.¹⁵ The most likely near-term role is as a decision-support layer that flags features, suggests a TI-RADS category, and lets the radiologist accept, adjust, or override the recommendation.

TIRADS in Children

Most TIRADS systems were designed and validated in adults, and their size thresholds do not translate cleanly to pediatric patients. A child’s thyroid is smaller, which means a nodule that seems small by adult standards can be proportionally large and clinically significant in a younger patient. Thyroid nodules in children also have a higher rate of malignancy compared to adults, roughly 20 to 25 percent versus around 5 to 15 percent, which changes the calculus on when to biopsy.

Researchers have begun developing pediatric-specific modifications. One multi-institutional study proposed a “PED TI-RADS” with adjusted size cutoffs: biopsy recommended for TR3 nodules at 1.5 centimeters (rather than 2.5), TR4 at 1.0 centimeter (rather than 1.5), and TR5 at any feasible size. Applying these thresholds, the researchers found that 88 patients could have been spared an unnecessary biopsy, with improved sensitivity and accuracy compared with existing guidelines.¹⁶ South Korea’s 2021 K-TIRADS revision also introduced pediatric-specific biopsy cutoffs.¹⁷ Another study found that while ACR TI-RADS maintained reasonable sensitivity in children and young adults (around 85 percent on average), its specificity dropped to about 51 percent, lower than the roughly 70 percent seen in adults.¹⁸ If your child has a thyroid nodule being evaluated, it is worth asking whether the size thresholds being applied are adult or pediatric.

When Background Thyroid Disease Complicates Things

Hashimoto’s thyroiditis, the most common cause of hypothyroidism, creates a lumpy, inflamed thyroid that can look suspicious on ultrasound even when no cancer is present. The chronic inflammation changes the texture of the gland and can mimic features that TIRADS systems use to flag malignancy, such as irregular margins and heterogeneous echogenicity. You might expect this to cause a flood of false positives, but the evidence is more nuanced.

A comparative study found that ACR TI-RADS maintained the lowest sensitivity but the highest specificity and the lowest rate of unnecessary biopsies in both patients with and without Hashimoto’s thyroiditis. Patients with Hashimoto’s actually had lower unnecessary biopsy rates overall, likely because the selection of nodules referred for evaluation in these patients is already somewhat filtered by clinical context.¹⁹ Still, certain features used for scoring, particularly margins, behaved differently in the Hashimoto’s group: the difference in margin appearance between benign and malignant nodules was significant in patients without Hashimoto’s but disappeared in those with it. Radiologists need to account for this when interpreting a TIRADS score in someone with known autoimmune thyroid disease.

Where TIRADS Meets Its Limits

TIRADS works well as a first-pass filter, but it does not answer every clinical question. One area where it falls short is in nodules that come back with indeterminate cytology after biopsy. These are cases where the biopsy was performed, the cells were examined under a microscope, and the pathologist could not definitively say benign or malignant. For these nodules, molecular testing of the biopsy sample has become increasingly important. A multicenter study found that once molecular testing results were available, neither the ATA ultrasound pattern nor the TI-RADS score added significant predictive value on top of the molecular result.²⁰ In other words, TIRADS is most useful before the biopsy, not after it. Once you have molecular data, the ultrasound appearance adds little new information.

Very small nodules present another challenge. Nodules under 1 centimeter are generally left alone under ACR TI-RADS unless they are TR5, but these tiny nodules are not always innocent. A narrative review noted that among referred subcentimeter nodules, the malignancy rate can actually be quite high because of selection bias: the ones that get referred tend to have been flagged for a reason. At the same time, most thyroid microcarcinomas found incidentally are clinically indolent and unlikely to cause harm, which is why current guidelines lean toward observation rather than intervention for the smallest nodules.²¹ Active surveillance with periodic ultrasound has emerged as a viable alternative to immediate surgery for small, low-risk papillary thyroid cancers, with studies showing low progression rates under monitoring.²²

The Psychological Side of a TIRADS Score

Something that does not show up in sensitivity and specificity calculations is how it feels to be told your thyroid nodule has been classified as “moderately suspicious” or “highly suspicious.” Many patients receive their TIRADS result through a radiology report or patient portal before their doctor has had a chance to explain it, and the language can be alarming. A TR4 or TR5 designation sounds like a cancer verdict to someone unfamiliar with the system, even though the vast majority of TR4 nodules turn out to be benign.

Research on the emotional impact has produced somewhat counterintuitive findings. A study of patients with TR4 nodules found that anxiety and depression scores had only weak correlations with the nodule’s size, TI-RADS level, or the presence of calcifications.²³ The imaging characteristics themselves were not driving the distress. Another study comparing patients whose nodules fell into higher versus lower risk categories found that those in the higher-risk group did report more somatization and anxiety symptoms overall, but the differences were modest.²⁴ What seems to matter more is uncertainty itself: not knowing whether a nodule is cancer, regardless of what the TIRADS category says, is what generates the most psychological burden. Clear communication from clinicians about what a TIRADS number actually means, and what the realistic probability of cancer is, can go a long way toward reducing unnecessary worry.