Comparative analysis of three-dimensional and two-dimensional models for predicting the malignancy probability of...

Area of Science:

• Pulmonary imaging and diagnostics
• Oncology
• Medical artificial intelligence

Background:

• Subsolid nodules (SSNs) require accurate malignancy prediction for optimal patient management.
• Current predictive models have limitations in distinguishing benign from malignant SSNs.

Purpose of the Study:

• To develop and compare the effectiveness of three-dimensional (3D) and two-dimensional (2D) models for predicting the malignancy probability of SSNs.
• To identify key factors contributing to accurate SSN malignancy prediction.

Main Methods:

• Construction of 3D and 2D predictive models using binary logistic backward regression on 371 SSNs from 332 patients.
• Validation of models using receiver operating characteristic (ROC) analysis and area under the curve (AUC) metrics.
• Comparison of model performance against the established Brock model using the DeLong test.

Main Results:

• In the training set, the two 3D models achieved AUCs of 0.785 and 0.776, outperforming the 2D model (0.764) and Brock model (0.738).
• In the test set, the 3D models yielded AUCs of 0.817 and 0.796, showing comparable performance to the Brock model (0.790) and 2D model (0.771).
• The 3D model incorporating volume, mean CT attenuation value, and lobulation demonstrated the highest predictive accuracy.

Conclusions:

• Three-dimensional models offer enhanced accuracy in predicting SSN malignancy compared to 2D and the Brock model.
• Specific features like volume, mean CT attenuation, and lobulation are crucial for improving 3D model performance.
• These findings support the potential of 3D modeling in refining SSN risk stratification and clinical decision-making.