A Novel Deep Learning Model to Distinguish Malignant Versus Benign Solid Lung Nodules.

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Distinguishing malignant from benign lung nodules is crucial for patient management.
• Current diagnostic methods can be invasive or lack definitive accuracy.
• Noncontrast and thin-layer computed tomography (CT) offer valuable imaging data.

Purpose of the Study:

• To develop and evaluate a novel transfer learning model for classifying solid lung nodules.
• To assess the performance of the Inception V3 deep learning model using CT scans.
• To compare the effectiveness of lesion-level versus image-level analysis.

Main Methods:

• Retrospective collection of CT images from 202 patients with histopathologically confirmed lung nodules.
• Application of the Inception V3 model, pre-trained on a training dataset, for nodule classification.
• Evaluation of model performance using receiver operator characteristic (ROC) curves, AUC, accuracy, sensitivity, and specificity.

Main Results:

• The Inception V3 model at the lesion-level achieved an AUC of 0.999, accuracy of 0.989, sensitivity of 0.983, and specificity of 1.0.
• Lesion-level analysis with Inception V3 demonstrated superior performance compared to image-level analysis.
• The Inception V3 model outperformed the ResNet50 model in differentiating lung nodules.

Conclusions:

• A novel deep learning model based on CT scans effectively classifies benign versus malignant lung nodules.
• The Inception V3 model significantly enhances differentiation accuracy and specificity.
• This approach offers a promising non-invasive tool for lung nodule characterization.