Patricia Sandqvist1,2, Anders Sundin3, Inga-Lena Nilsson1,4, Per Grybäck1,2, Alejandro Sanchez-Crespo2,5. 1. Department of Molecular Medicine and Surgery, Karolinska Institute, Stockholm, Sweden. 2. Department of Medical Radiation Physics and Nuclear Medicine, Karolinska University Hospital, Stockholm, Sweden. 3. Department of Radiology, Section for Molecular Imaging, Institution for Surgical Sciences, Uppsala University Hospital, Uppsala, Sweden. 4. Department of Breast, Endocrine Tumours and Sarcoma, Karolinska University Hospital, Stockholm, Sweden. 5. Department of Oncology-Pathology, Karolinska Institute, Stockholm, Sweden.
Abstract
Objective: Successful preoperative image localisation of all parathyroid adenomas (PTA) in patients with primary hyperparathyroidism (pHPT) and multiglandular disease (MGD) remains challenging. We investigate whether a machine learning classifier (MLC) could predict the presence of overlooked PTA at preoperative localisation with 99mTc-Sestamibi-SPECT/CT in MGD patients. Design: This study is a retrospective study from a single tertiary referral hospital initially including 349 patients with biochemically confirmed pHPT and cured after surgical parathyroidectomy. Methods: A classification ensemble of decision trees with Bayesian hyperparameter optimisation and five-fold cross-validation was trained with six predictor variables: the preoperative plasma concentrations of parathyroid hormone, total calcium and thyroid-stimulating hormone, the serum concentration of ionised calcium, the 24-h urine calcium and the histopathological weight of the localised PTA at imaging. Two response classes were defined: patients with single-gland disease (SGD) correctly localised at imaging and MGD patients in whom only one PTA was localised on imaging. The data set was split into 70% for training and 30% for testing. The MLC was also tested on a subset of the original data based on CT image-derived PTA weights. Results: The MLC achieved an overall accuracy at validation of 90% with an area under the cross-validation receiver operating characteristic curve of 0.9. On test data, the MLC reached a 72% true-positive prediction rate for MGD patients and a misclassification rate of 6% for SGD patients. Similar results were obtained in the testing set with image-derived PTA weight. Conclusions: Artificial intelligence can aid in identifying patients with MGD for whom 99mTc-Sestamibi-SPECT/CT failed to visualise all PTAs.
Objective: Successful preoperative image localisation of all parathyroid adenomas (PTA) in patients with primary hyperparathyroidism (pHPT) and multiglandular disease (MGD) remains challenging. We investigate whether a machine learning classifier (MLC) could predict the presence of overlooked PTA at preoperative localisation with 99mTc-Sestamibi-SPECT/CT in MGD patients. Design: This study is a retrospective study from a single tertiary referral hospital initially including 349 patients with biochemically confirmed pHPT and cured after surgical parathyroidectomy. Methods: A classification ensemble of decision trees with Bayesian hyperparameter optimisation and five-fold cross-validation was trained with six predictor variables: the preoperative plasma concentrations of parathyroid hormone, total calcium and thyroid-stimulating hormone, the serum concentration of ionised calcium, the 24-h urine calcium and the histopathological weight of the localised PTA at imaging. Two response classes were defined: patients with single-gland disease (SGD) correctly localised at imaging and MGD patients in whom only one PTA was localised on imaging. The data set was split into 70% for training and 30% for testing. The MLC was also tested on a subset of the original data based on CT image-derived PTA weights. Results: The MLC achieved an overall accuracy at validation of 90% with an area under the cross-validation receiver operating characteristic curve of 0.9. On test data, the MLC reached a 72% true-positive prediction rate for MGD patients and a misclassification rate of 6% for SGD patients. Similar results were obtained in the testing set with image-derived PTA weight. Conclusions: Artificial intelligence can aid in identifying patients with MGD for whom 99mTc-Sestamibi-SPECT/CT failed to visualise all PTAs.