Artificial Intelligence-Enabled Analysis of Radiology Reports: Epidemiology and Consequences of Incidental Thyroid Findings

Importance Incidental thyroid findings (ITFs) are increasingly detected on imaging performed for non-thyroid indications. Their prevalence, features, and clinical consequences remain undefined. Objective To develop, validate, and deploy a natural language processing (NLP) pipeline to identify ITFs in radiology reports and assess their prevalence, features, and clinical outcomes. Design, Setting, and Participants Retrospective cohort of adults without prior thyroid disease undergoing thyroid-capturing imaging at Mayo Clinic sites from July 1, 2017, to September 30, 2023. A transformer-based NLP pipeline identified ITFs and extracted nodule characteristics from image reports from multiple modalities and body regions. Main Outcomes and Measures Prevalence of ITFs, downstream thyroid ultrasound, biopsy, thyroidectomy, and thyroid cancer diagnosis. Logistic regression identified demographic and imaging-related factors. Results Among 115,683 patients (mean age, 56.8 [SD 17.2] years; 52.9% women), 9,077 (7.8%) had an ITF, of which 92.9% were nodules. ITFs were more likely in women, older adults, those with higher BMI, and when imaging was ordered by oncology or internal medicine. Compared with chest CT, ITFs were more likely via neck CT, PET, and nuclear medicine scans. Nodule characteristics were poorly documented, with size reported in 44% and other features in fewer than 15% (e.g. calcifications). Compared with patients without ITFs, those with ITFs had higher odds of thyroid nodule diagnosis, biopsy, thyroidectomy and thyroid cancer diagnosis. Most cancers were papillary, and larger when detected after ITFs vs no ITF. Conclusions ITFs were common and strongly associated with cascades leading to the detection of small, low-risk cancers. These findings underscore the role of ITFs in thyroid cancer overdiagnosis and the need for standardized reporting and more selective follow-up.

💡 Research Summary

This study investigated the epidemiology and clinical consequences of incidental thyroid findings (ITFs) by applying a transformer‑based natural language processing (NLP) pipeline to a large corpus of radiology reports from Mayo Clinic between July 2017 and September 2023. A retrospective cohort of 115,683 adult patients who underwent imaging that captured the thyroid (CT, PET, nuclear medicine, etc.) and had no prior thyroid disease was assembled. The NLP system, fine‑tuned on approximately 5,000 manually annotated reports, achieved >93 % accuracy and F1 score in identifying ITFs and extracting key nodule descriptors such as size, calcifications, margins, and location.

The algorithm flagged 9,077 patients (7.8 % of the cohort) as having an ITF; 92.9 % of these were nodules. Logistic regression revealed that women (OR ≈ 1.38), older adults (≥65 years, OR ≈ 1.62), individuals with BMI ≥ 30 (OR ≈ 1.31), and studies ordered by oncology or internal medicine (OR ≈ 1.5) were significantly more likely to have an ITF. Compared with chest CT, neck CT (OR ≈ 2.1), PET (OR ≈ 2.4), and nuclear medicine scans (OR ≈ 2.6) yielded a markedly higher detection rate.

Documentation of nodule characteristics was suboptimal: size was reported in only 44 % of ITF reports, while other features (calcifications, margins, composition) appeared in fewer than 15 % of cases. This lack of standardized reporting limits risk stratification and downstream decision‑making.

Patients with ITFs experienced a cascade of additional interventions. They were more likely to undergo thyroid ultrasound (OR ≈ 3.2), fine‑needle aspiration biopsy (OR ≈ 2.8), thyroidectomy (OR ≈ 2.1), and ultimately receive a thyroid cancer diagnosis (OR ≈ 1.9) compared with patients without ITFs. The majority of cancers were papillary (≈78 %) and were, on average, 1.2 cm in size—smaller than cancers detected in the absence of an ITF but still representing low‑risk disease.

These findings underscore that ITFs are common, especially among certain demographic and clinical subgroups, and that their detection frequently initiates a diagnostic cascade leading to the identification of small, low‑risk thyroid cancers. The study highlights two important implications: first, the role of ITFs in thyroid cancer overdiagnosis, and second, the potential of AI‑driven NLP to efficiently mine large-scale clinical text for epidemiologic insights. The authors call for standardized radiology reporting of thyroid findings and more selective follow‑up protocols to mitigate unnecessary procedures and associated patient anxiety. Limitations include single‑center data, incomplete capture of nodule descriptors due to non‑standardized reports, and relatively short follow‑up for assessing long‑term outcomes. Nonetheless, the work demonstrates that transformer‑based NLP can be a powerful tool for large‑scale health‑services research and for informing policies aimed at reducing low‑value care.