A visual deep learning model to predict abnormal versus normal parathyroid glands using intraoperative...

Area of Science:

• Endocrinology
• Surgical Oncology
• Artificial Intelligence in Medicine

Background:

• Abnormal parathyroid glands (PGs) show distinct autofluorescence patterns compared to normal glands.
• Previous research indicated darker, more heterogeneous autofluorescence in abnormal PGs.
• Intraoperative assessment of PGs is crucial during surgery for primary hyperparathyroidism.

Purpose of the Study:

• To develop a visual artificial intelligence (AI) model for predicting parathyroid gland abnormality.
• To utilize intraoperative autofluorescence signals to differentiate between normal and abnormal PGs.
• To assess the AI model's efficacy in identifying hypersecreting and/or hypercellular PGs before excision.

Main Methods:

• A dataset of 906 intraoperative parathyroid autofluorescence images from 303 patients was used.
• Images were uploaded to a visual AI platform, labeled as abnormal or normal.
• A deep learning approach with 80% training, 10% testing, and 10% validation data was employed.

Main Results:

• The AI model achieved an area under the receiver operating characteristic curve (AUROC) of 0.90.
• The area under the precision-recall curve (AUPRC) for the model was 0.93.
• The model demonstrated 89% recall (sensitivity) and 89% precision (positive predictive value).

Conclusions:

• Visual AI platforms can effectively compare intraoperative autofluorescence signals against extensive databases.
• This AI model shows potential as an adjunctive tool for intraoperative parathyroid gland assessment.
• The technology may improve surgical decision-making during exploration for primary hyperparathyroidism.