Intraoperative parathyroid gland recognition prediction model and key feature analysis based on white light images.

Abstract

Background: Timely identification and protection of the parathyroid glands during thyroid surgery are important. This study aims to explore a convenient, efficient, and inexpensive method for identifying parathyroid glands during surgery by extracting subtle texture features that cannot be recognized by the naked eye from white light images.

Methods: In total, 117 confirmed parathyroid gland photos and 169 oval tissue photos of non-parathyroid glands, such as suspected parathyroid gland fat granules and lymph nodes, were collected. All the photos were subjected to color channel conversion, a region of interest (ROI) was drawn, and seven major types of texture features were extracted for each color channel using the PyRadiomics package. The least absolute shrinkage and selection operator (LASSO) algorithm was used to screen key features, and multiple machine learning algorithms were used to establish a prediction model on the basis of the above texture features. The SHapley Additive exPlanations (SHAP) algorithm was applied for key feature analysis.

Results: A parathyroid gland prediction model based on white light texture features was successfully established, with the best performance achieved using the random forest (RF) algorithm. The accuracy, specificity, sensitivity, and area under the receiver operating characteristic (ROC) curve were 89.6%, 85.7%, 91.8%, 88.7%, and 77.5%, respectively. The SHAP algorithm revealed several key texture features of the parathyroid gland.

Conclusions: This study is the first to establish and validate a convenient and economical intraoperative parathyroid gland identification model, which has potential clinical application value.