Yeh Randy, Kuo Jennifer H, Huang Bernice, Shobeiri Parnian, Lee James A, Tay Yu-Kwang Donovan, Tabacco Gaia, Bilezikian John P, Dercle Laurent
Department of Radiology, New York-Presbyterian Hospital/Columbia University Medical Center, New York, NY, USA.
Department of Radiology, Molecular Imaging and Therapy Service, Memorial Sloan Kettering Cancer Center, New York, NY, USA.
Eur Radiol. 2025 Mar;35(3):1325-1336. doi: 10.1007/s00330-024-11159-8. Epub 2024 Oct 30.
To train and validate machine learning-derived clinical decision algorithm (CDA) for the diagnosis of hyperfunctioning parathyroid glands using preoperative variables to facilitate surgical planning.
This retrospective study included 458 consecutive primary hyperparathyroidism (PHPT) patients who underwent combined 4D-CT and sestamibi SPECT/CT (MIBI) with subsequent parathyroidectomy from February 2013 to September 2016. The study cohort was divided into training (first 400 patients) and validation sets (remaining 58 patients). Sixteen clinical, laboratory, and imaging variables were evaluated. A random forest algorithm selected the best predictor variables and generated a clinical decision algorithm with the highest performance (CDA). The CDA was trained to predict the probability of a hyperfunctioning vs normal gland for each of the four parathyroid glands in a patient. The reference standard was a four-quadrant location on operative reports and pathology. The accuracy of CDA was prospectively validated.
Of 16 variables, the algorithm selected 3 variables for optimal prediction: combined 4D-CT and MIBI using (1) sensitive reading, (2) specific reading, and (3) cross-product of serum calcium and parathyroid hormone levels and outputted an CDA using five probability categories for hyperfunctioning glands. The CDA demonstrated excellent accuracy for correct classification in the training (4D-CT + MIBI: 0.91 [95% CI: 0.89-0.92]) and validation sets (4D-CT + MIBI: 0.90 [95% CI: 0.86-0.94].
Machine learning generated a clinical decision algorithm that accurately diagnosed hyperfunctioning parathyroid glands through classification into probability categories, which can be implemented for improved preoperative planning and convey diagnostic certainty.
Question Can an CDA use preoperative variables for the diagnosis of hyperfunctioning parathyroid glands to facilitate surgical planning? Findings The developed CDA demonstrated excellent accuracy for correct classification in the training (0.91 [95% CI: 0.89-0.92]) and validation sets (0.90 [95% CI: 0.86-0.94]). Clinical relevance Using standard preoperative variables, an CDA for diagnosing hyperfunctioning parathyroid glands can be implemented to improve preoperative parathyroid localization and included in radiology reports for surgical planning.
训练并验证基于机器学习的临床决策算法（CDA），该算法利用术前变量诊断甲状旁腺功能亢进，以辅助手术规划。
这项回顾性研究纳入了2013年2月至2016年9月期间连续458例接受了4D-CT与甲氧基异丁基异腈单光子发射计算机断层扫描/计算机断层扫描（MIBI）联合检查并随后接受甲状旁腺切除术的原发性甲状旁腺功能亢进（PHPT）患者。研究队列分为训练组（前400例患者）和验证组（其余58例患者）。评估了16项临床、实验室和影像学变量。随机森林算法选择了最佳预测变量，并生成了性能最佳的临床决策算法（CDA）。训练CDA以预测患者四个甲状旁腺中每个甲状旁腺功能亢进与正常的概率。参考标准为手术报告和病理中的四个象限位置。前瞻性地验证了CDA的准确性。
在16项变量中，该算法选择了3项变量进行最佳预测：联合使用（1）敏感读数、（2）特异读数的4D-CT和MIBI，以及（3）血清钙和甲状旁腺激素水平的交叉乘积，并使用针对功能亢进腺体的五个概率类别输出了一个CDA。CDA在训练组（4D-CT + MIBI：0.91 [95% CI：0.89 - 0.92]）和验证组（4D-CT + MIBI：0.90 [95% CI：0.86 - 0.94]）中对正确分类显示出优异的准确性。
机器学习生成了一种临床决策算法，该算法通过分类为概率类别准确诊断甲状旁腺功能亢进，可用于改进术前规划并传达诊断确定性。
问题 一种CDA能否使用术前变量诊断甲状旁腺功能亢进以辅助手术规划？ 发现 所开发的CDA在训练组（0.91 [95% CI：0.89 - 0.92]）和验证组（0.90 [95% CI：0.86 - 0.94]）中对正确分类显示出优异的准确性。 临床意义 使用标准术前变量，可实施一种用于诊断甲状旁腺功能亢进的CDA，以改善术前甲状旁腺定位，并纳入放射学报告以用于手术规划。
