‘Returning to TI-RADS’ may assist with triage of indeterminate thyroid nodules
A re-analysis of thyroid imaging reporting and data system ultrasound scoring after molecular analysis is a cost-effective option to assist with preoperative diagnosis of indeterminate thyroid nodules, data from a retrospective study show.
Indeterminate thyroid nodules — those with cytopathology classified as Bethesda III from a scale of I to VI — are challenging to characterize without diagnostic surgery, Matti Gild, MBBS (Hons), PhD, FRACP, an endocrinologist at Royal North Shore Hospital, Sydney, and senior lecturer in medicine at the University of Sydney, and colleagues wrote in Clinical Endocrinology. Auxiliary strategies, such as molecular analysis, machine learning models and ultrasound grading with the thyroid imaging reporting and data system (TI‐RADS) scoring system, can help triage accordingly; however, further refinement is needed to prevent unnecessary surgeries and increase positive predictive values.
“We can apply a more personalized approach to follow indeterminate thyroid nodules depending on TI-RADS score and patient age rather than a generic ‘repeat biopsy in 3 months,’ which guidelines recommend,” Gild told Healio. “This study shows that novel radiomic and radiologic strategies can be employed to assist with preoperative diagnosis of indeterminate thyroid nodules.”
Re-stratifying risk
In a retrospective review, Gild and colleagues analyzed data from 88 patients with Bethesda III nodules who went on to have diagnostic surgery between 2014 and 2019 with final pathological diagnosis (mean age, 60 years). Histology from the malignant nodules included 20 classified as papillary thyroid cancer, one as insular carcinoma, seven as follicular thyroid cancer, three Hurthle cell cancers and three classified as other.
Each nodule was retrospectively scored through TI‐RADS, and two deep learning models were tested. One, ThyNet, was previously developed and trained on another data set, mainly containing determinate cases; a second model trained and was tested on the indeterminate cases reviewed by the researchers.
Within the cohort, the mean TI‐RADS score was 3 for benign and 4 for malignant nodules (P = .0022).
Researchers stratified nodules as radiological high risk (TI‐RADS 4 or 5) and low risk (TI‐RADS 2 or 3). The positive predictive value for the high radiological risk category among patients with nodules greater than 10 mm (n = 44) was 85% (95% CI, 70-93).
The negative predictive value for low radiological risk in patients aged at least 60 years was 100% (95% CI, 83-100).
“Older patients with indeterminate nodules may be reassured that the likelihood of a final malignant pathology is low, and they may feel more confident in observation rather than diagnostic surgery,” Gild told Healio.
The researchers noted molecular testing is costly, often not subsidized and not easily accessible to those in less resourced countries; using the validated TI‐RADS score to re-stratify risk is a “simple, cost-neutral first‐line approach.”
“In a cohort of suspicious lesions from molecular analysis, returning to TI‐RADS grade may assist with triage and surgical timing,” the researchers wrote.
More machine learning data needed
The area under the curve value of the novel classifier was 0.75 (95% CI, 0.62-0.84) and differed significantly from the chance‐level (P < .00001), according to the researchers.
Gild said additional analyses with larger data sets may yield more meaningful findings regarding the utility of machine learning in predicting nodule malignancy.
“The AUC through the deep learning models designed here were not high enough to be clinically meaningful; however, these methods are novel,” Gild told Healio. “Further refinement and larger data sets may assist in characterization of these nodules prior to surgery.”
For more information:
Matti Gild, MBBS (Hons), PhD, FRACP, can be reached at matti.gild@sydney.edu.au; Twitter: @mattilg.
Collapse
Read more about