# Reliability of Ultrasound Features in Predicting Thyroid Malignancy

## TL;DR

This study evaluates how well specific ultrasound features can predict thyroid cancer, finding that a combination of two features offers a good balance between accuracy and reliability.

## Contribution

The study introduces a predefined five-feature ultrasound composite and identifies a two-feature subset (irregular margin and microcalcification) for improved diagnostic performance.

## Key findings

- A predefined five-feature ultrasound composite showed 95.45% sensitivity in detecting thyroid malignancy.
- The combination of irregular margin and microcalcification provided a balanced sensitivity-specificity trade-off (74.4% and 73.5%, respectively).
- Ultrasound criteria showed strong diagnostic performance but should be used alongside other clinical assessments for management decisions.

## Abstract

Introduction: Thyroid nodules have shown an increasing trend over the past years, with a recent surge. Ultrasound is the primary modality of investigation to detect malignancy in thyroid nodules, which is safe, cost-effective, and easily accessible, whereas fine needle aspiration biopsy is an invasive and costly test with risk factors. This study aims to assess how accurately a predefined five-feature ultrasound (US) composite (microcalcifications, marked hypo echogenicity, irregular/micro lobulated margins, intranodular vascularity, and suspicious lymph nodes) can identify disease compared to histopathology by measuring its sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC) with 95% confidence intervals. The ultrasound composite is defined a priori as positive if two or more of the five features are present. The secondary objective is to explore which smaller combination of features provides the best diagnostic performance based on ROC analysis.

Methodology: Between August 2015 and December 2022, 313 consecutive patients with thyroid nodules presenting to the Surgical Department were screened; out of these, 146 patients underwent surgery for the presence of large size, suspicious swelling, and other cosmetic reasons. All patients underwent routine preoperative fine needle aspiration cytology (FNAC) and neck ultrasound. Final histopathology comprised the diagnostic accuracy analysis set (per-nodule analysis of the surgically excised nodule). Ultrasound features and FNAC results were correlated with the final histopathological report. US features recorded were hypoechogenicity, irregular margin, microcalcification, increased vascularity, and a suspicious lymph node. Data was statistically analyzed using the multiple logistic regression method, the receiver operating characteristic (ROC) curve, the cross-tabulation McNemar’s chi-square test, and IBM Corp. Released 2012. IBM SPSS Statistics for Windows, Version 18. Armonk, NY: IBM Corp. We prespecified the diagnostic rule a priori as US-positive if ≥2 of the five features were present. Diagnostic performance measures were calculated against histopathology.

Results: Using ultrasound diagnostic criteria, 112 patients were classified as positive, and histopathology later confirmed that 84 of these had malignancies. This showed a sensitivity of 95.45%, a specificity of 51%, and a positive predictive value of 75%. Further analysis via the ROC curve showed a high malignancy probability at a point that corresponds to the presence of the 2 ultrasound characteristics, i.e., irregular margin and microcalcification. Specificity and sensitivity at this point are 73.5 and 74.4, respectively.

Conclusion: The predefined rule using ≥2 features showed high sensitivity for detecting malignant thyroid nodules, making it useful for screening. The presence of irregular margins and microcalcifications (M + C pair) provided a more balanced sensitivity-specificity trade-off, supporting its use for risk stratification rather than direct surgical decision-making. While these ultrasound criteria show strong diagnostic performance, definitive management decisions should continue to rely on multidisciplinary assessment, integrating cytology, clinical findings, and patient factors. Further prospective validation in larger, unselected cohorts is recommended to refine threshold criteria and confirm generalizability.