[The diagnostic value of machine-learning-based model for predicting the malignancy of solid nodules in multiple...

Area of Science:

• Artificial Intelligence in Medicine
• Pulmonary Nodule Diagnosis
• Machine Learning for Healthcare

Background:

• Accurate diagnosis of solid pulmonary nodules is crucial for patient management.
• Existing diagnostic models often struggle with the complexity of multiple pulmonary nodules.
• Machine learning offers potential for improved diagnostic accuracy by integrating diverse data.

Purpose of the Study:

• To develop and evaluate a machine learning diagnostic model (PKU-ML) for solid pulmonary nodules.
• To assess the model's efficacy specifically in cases of multiple pulmonary nodules.
• To compare the PKU-ML model's performance against established diagnostic models.

Main Methods:

• Utilized extreme gradient boosting (XGBoost) algorithm on a dataset of 446 solid nodules from 287 patients.
• Combined patient clinical information and CT features for model construction.
• Validated the model on separate training, test, and independent single nodule datasets using Area Under the Curve (AUC).

Main Results:

• The PKU-ML model achieved high diagnostic accuracy in the training set (AUC=0.883) and test set (AUC=0.838).
• PKU-ML demonstrated superior performance compared to Brock, Mayo, and VA models for multiple pulmonary nodules.
• The model also showed good predictive value for single solid pulmonary nodules (AUC=0.786).

Conclusions:

• The machine learning-based PKU-ML model significantly enhances the prediction of malignancy in solid nodules within multiple pulmonary nodule cases.
• PKU-ML offers a robust diagnostic tool, performing well even for single solid pulmonary nodules.
• This AI-driven approach represents a valuable advancement over traditional mathematical models for pulmonary nodule diagnosis.