CT Radiomics Model Without Contrast Enhances Detection of PTC in Hashimoto's Thyroiditis

Overview

This study developed a nonenhanced CT (NECT) radiomics model to improve detection of papillary thyroid carcinoma (PTC) in patients with Hashimoto's thyroiditis (HT). The model demonstrated superior diagnostic performance compared to conventional imaging, addressing challenges in identifying occult PTC masked by HT-related thyroid changes.

Background

Hashimoto’s thyroiditis is a common benign thyroid disorder characterized by chronic inflammation and tissue destruction, often coexisting with papillary thyroid carcinoma. The incidence of PTC is higher in HT patients, complicating diagnosis due to overlapping imaging features. Current diagnostic standards rely on invasive biopsy, which can miss occult PTCs. Radiomics, extracting quantitative imaging features for machine learning analysis, offers a promising noninvasive approach to detect hidden malignancies within the inflammatory thyroid environment.

Data Highlights

Key Findings

• A total of 130 patients with pathologically confirmed HT and preoperative NECT were included after exclusions.
• 851 radiomic features were extracted per patient from delineated thyroid regions on NECT images.
• Feature selection involved ICC filtering (≥0.75), Pearson correlation analysis, and LASSO regression to reduce dimensionality.
• Four machine learning models (logistic regression, naive Bayes, support vector machine, multilayer perceptron) were constructed using selected features.
• Model performance was evaluated by ROC curve analysis, with AUC as the primary metric, and decision curve analysis assessed clinical utility.
• SHAP analysis identified the most influential radiomic features contributing to the best-performing model.

Clinical Implications

The NECT-based radiomics model provides a noninvasive tool to improve detection of occult PTC in patients with HT, potentially reducing reliance on invasive biopsy and minimizing missed diagnoses. This approach can aid clinicians in timely identification and management of malignancy within the challenging inflammatory background of HT. Incorporation of radiomics into routine imaging protocols may enhance diagnostic accuracy and guide surgical decision-making.

Conclusion

A radiomics model derived from nonenhanced CT images significantly improves detection of papillary thyroid carcinoma in patients with Hashimoto’s thyroiditis. This technique offers a promising adjunct to conventional imaging and biopsy, facilitating earlier and more accurate diagnosis.

References

1. Fang et al. 2021 -- Ultrasound Radiomics in Thyroid Nodules with Hashimoto's Thyroiditis
2. Institutional Research Ethics Board Approval -- Retrospective Study on HT and PTC