Advanced Thyroid Imaging May Aid Diagnosis of Indeterminate Thyroid Nodules

Fine needle biopsy is the standard diagnostic test for indeterminate thyroid nodules, but increases medical costs and causes anxiety in patients. Researchers explored whether targeted use of Thyroid Imaging, Reporting, and Data System (TI-RADS) could provide faster and less-invasive diagnoses.

The Thyroid Imaging, Reporting, and Data System (TI‐RADS) may be an alternative to more invasive diagnostic measures for thyroid nodules, according to a study in Clinical Endocrinology.

Researchers evaluated how TI‐RADS analysis and machine learning models could optimize risk stratification in indeterminate thyroid nodules. They used 2 different deep‐learning approaches. In the first approach, images were fed directly as inputs of the deep learning–based classifier. In the second approach, the nodules were segmented with use of a deep learning–based model, and histogram‐based and textural radiomic features were extracted from each segmented area.

A total of 88 patients (mean age, 60.1 years [range, 24-88 years]) with Bethesda III nodules were included. The median size of the benign nodules was 24 mm (range, 6‐45 mm) vs 22.5 mm (range, 4.5-80 mm) for malignant nodules, with no statistically significant difference (P = .87).

A majority of nodules were classified as TI‐RADS 4 or ‘moderately suspicious’ (37, 42%), followed by TI‐RADS 3 or ‘mildly suspicious’ (30, 34%), TI‐RADS 5 or ‘highly suspicious’ (14, 16%), and TI‐RADS 2 or ‘not suspicious’ (7, 8%).

The mean TI‐RADS score for benign pathology was 3.4 (median 3). The mean TI‐RADS score for final malignant pathology was 4.0 (median 4). The Mann–Whitney test was used to determine a difference between the median scores, with a statistically significant difference found in the cohorts (P = .0022).

In patients aged >60 years, 57% (27/47) of the nodules were rated in a high radiological risk category with a 23% malignancy risk. For patients ˂60 years, the disease prevalence was higher (56%) but in a similar proportion (56%; 24/41) were graded TI-RADS 4 or 5. In these cohorts, the positive predictive values (PPV) were 40% (CI, 32%-49%) and 67% (CI, 53%-78%), respectively. The negative predictive value (NPV) for the ˃60 cohort was 100% and ˂60 was 59% (CI, 40%-65%).

Researchers evaluated whether higher radiological risk category predicted malignancy in larger nodules. The PPV for nodules >10 mm was 85% (CI, 70%-93%), and NPV was 67% (CI, 48%-81%) in the >10 mm group. Findings for nodules <10 mm were too small to be clinically meaningful.

An area under the receiver operating characteristic curve of 0.75 (CI, 0.62-0.84) was achieved after all first‐order statistics radiomic features were fed into a random forest classifier.

“The difference in median TI‐RADS scores for the benign and malignant nodules was statistically significant, demonstrating the utility of this measure in our cohort,” stated the researchers.

“While there is little doubt that refinement of next generation sequencing panels have improved diagnostic capabilities and thus led to less unnecessary diagnostic surgery, cost and availability [are] prohibitive for lower resourced regions,” the study authors commented. “Here we have presented alternative options, which under further refinement may be more accessible and cost effective.”


Gild ML, Chan M, Gajera J, Lurie, B, Gandomkar, Z, Clifton-Bligh, RJ.  Risk stratification of indeterminate thyroid nodules using ultrasound and machine learning algorithms. Clin Endocrinol (Oxf). Published online October 13, 2021. doi: 10.1111/cen.14612