[Computer aided diagnosis model for lung tumor based on ensemble convolutional neural network].

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Cancer Detection
• Radiomics and Quantitative Imaging Analysis

Context:

• Lung tumors are often diagnosed using positron emission tomography (PET)/computed tomography (CT) imaging.
• Accurate quantitative analysis is crucial for lung tumor diagnosis, yet visual interpretation has limitations.
• Computer-aided diagnosis (CAD) systems aim to enhance diagnostic accuracy and efficiency.

Purpose:

• To develop and evaluate an ensemble convolutional neural network (CNN) for lung tumor recognition in PET/CT images.
• To compare the diagnostic performance of the ensemble CNN against individual CNN models (CT-CNN, PET-CNN, PET/CT-CNN).
• To investigate the influence of model parameters (epochs, batch size, image scale) on CNN training for lung tumor recognition.

Summary:

• Three CNNs (CT-CNN, PET-CNN, PET/CT-CNN) were constructed using a parameter migration method for lung tumor recognition in CT, PET, and PET/CT images.
• Model parameters for CT-CNN were optimized by analyzing their impact on recognition rate and training time.
• An ensemble CNN was created by combining the three single CNNs, with lung tumor recognition performed via a relative majority vote; the ensemble CNN outperformed single CNNs.

Impact:

• The developed ensemble CNN shows improved accuracy in computer-aided diagnosis of lung tumors compared to individual CNN models.
• This approach offers a more robust and accurate quantitative analysis for lung tumor detection, potentially aiding clinical decision-making.
• The findings suggest that ensemble deep learning models can effectively overcome limitations of single models in medical image analysis.