Detection of Lung Nodules in Micro-CT Imaging Using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Preclinical Research

Background:

• Co-clinical trials require efficient methods to monitor treatment response, particularly for lung metastasis.
• Micro-computed tomography (micro-CT) is used for longitudinal imaging of lung nodules in mouse models.
• Automated tumor detection can accelerate image analysis in preclinical studies.

Purpose of the Study:

• To explore deep learning (DL) for automated lung nodule detection in micro-CT scans of mice.
• To evaluate the impact of different training data strategies on DL model performance.
• To assess the feasibility of DL for monitoring lung metastasis in immunotherapy and radiotherapy co-clinical trials.

Main Methods:

• Longitudinal micro-CT imaging of mice with and without lung tumors.
• Data augmentation using simulated tumors inserted into real micro-CT scans.
• Training convolutional neural networks (CNNs) with four distinct datasets: simulated only, real only, combined, and pretraining on simulated then real data.
• Performance evaluation using precision-recall curves, ROC curves, and AUC.

Main Results:

• All four training strategies yielded similar Area Under the Curve (AUC) values (0.76-0.77).
• Combining real and simulated data improved detection precision by 8% compared to other methods.
• Detection rates were lower for smaller tumors, with networks trained on real data showing superior performance.
• Deep learning demonstrated a promising approach for rapid and accurate lung tumor detection in mice.

Conclusions:

• Deep learning offers a viable and efficient method for automated lung nodule detection in preclinical micro-CT imaging.
• Hybrid training datasets (real and simulated) enhance the precision of DL-based tumor detection.
• DL models show potential for accelerating the analysis of lung metastasis in mouse models relevant to co-clinical trials.