Machine learning-derived clinical decision algorithm for the diagnosis of hyperfunctioning parathyroid glands in...

Area of Science:

• Endocrinology
• Medical Imaging
• Machine Learning

Background:

• Primary hyperparathyroidism (PHPT) management relies on accurate preoperative localization of hyperfunctioning parathyroid glands.
• Current localization methods can be limited, impacting surgical planning and outcomes.

Purpose of the Study:

• To develop and validate a machine learning-derived clinical decision algorithm (MLCDA) for diagnosing hyperfunctioning parathyroid glands.
• To utilize preoperative variables for enhanced surgical planning in PHPT patients.

Main Methods:

• A retrospective study of 458 PHPT patients who underwent 4D-CT and sestamibi SPECT/CT (MIBI).
• A random forest algorithm was employed to select optimal predictor variables from 16 evaluated parameters.
• The MLCDA was trained to predict the probability of hyperfunctioning glands based on combined imaging and laboratory data.

Main Results:

• The MLCDA identified three key predictors: sensitive and specific readings from 4D-CT/MIBI, and the product of serum calcium and parathyroid hormone levels.
• The algorithm achieved excellent accuracy in both training (0.91) and validation (0.90) sets.
• The MLCDA classified gland function into five probability categories, demonstrating high diagnostic performance.

Conclusions:

• Machine learning can generate a robust clinical decision algorithm for diagnosing hyperfunctioning parathyroid glands.
• The developed MLCDA accurately predicts gland status, facilitating improved preoperative localization and surgical planning.
• This tool can enhance diagnostic certainty and be integrated into radiology reports for surgical guidance.