Classification of Pulmonary Nodules in 2-[¹⁸F]FDG PET/CT Images with a 3D Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• 2-[¹⁸F]FDG PET/CT is crucial for pulmonary nodule management.
• Convolutional neural networks (CNNs) can potentially enhance the differentiation between malignant and benign pulmonary nodules by automatically learning image features.

Purpose of the Study:

• To develop and validate a CNN model for classifying pulmonary nodules using 2-[¹⁸F]FDG PET images.

Main Methods:

• Retrospective analysis of 113 participants' 2-[¹⁸F]FDG PET images.
• Training and evaluation of three 3D CNN architectures (Stacked 3D CNN, VGG-like, Inception-v2-like) and transfer learning with ResNet-50.
• Four-fold cross-validation and testing on a held-out set.

Main Results:

• The final Stacked 3D CNN model achieved an Area Under the ROC Curve of 0.8385.
• The model demonstrated a sensitivity of 80.00%, specificity of 69.23%, and accuracy of 73.91% on the test set.
• An optimized decision threshold was used, prioritizing the reduction of false negatives.

Conclusions:

• A 3D CNN model proved effective in differentiating benign from malignant pulmonary nodules on 2-[¹⁸F]FDG PET images.
• This AI-driven approach holds potential for improving the diagnostic accuracy in the clinical management of pulmonary nodules.