Predicting Malignancy in Pediatric Thyroid Nodules: Early Experience With Machine Learning for Clinical Decision...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Pediatric Endocrinology

Background:

• Papillary thyroid carcinoma is the most common endocrine malignancy.
• Distinguishing benign from malignant thyroid nodules is crucial for patient management.
• Current diagnostic challenges include identifying malignancy while avoiding surgical risks for benign nodules.

Purpose of the Study:

• To develop an interpretable machine learning tool for predicting non-benign thyroid cytology and histology.
• Integrate clinical, demographic, ultrasound, and biopsy data for enhanced prediction accuracy.

Main Methods:

• Utilized Random Forests with feature selection on patient data (under 18 years).
• Incorporated interpretable rule sets for enhanced clinical decision-making.
• Evaluated model performance using accuracy, false-positive rate (FPR), false-negative rate (FNR), and area under the receiver operator curve (AUROC).

Main Results:

• Models demonstrated improved prediction of non-benign cytology and malignant histology compared to historical data.
• Projected improvements include a 11.90% increase in accuracy and 24.85% increase in AUROC for biopsy predictions.
• Projected improvements include a 32.19% increase in accuracy and a 68.04% decrease in FPR for surgery predictions.

Conclusions:

• This study presents a novel interpretable machine learning tool for thyroid nodule assessment.
• The tool aims to assist clinicians in identifying potentially malignant nodules more effectively.
• Future research will focus on expanding the dataset and developing probabilistic predictive estimates.