Lung nodule segmentation via semi-residual multi-resolution neural networks

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Cloud Computing

Background:

• Deep neural networks and cloud computing are increasingly integrated into medical image processing.
• Advancements in neural network theory and the Internet of Things (IoT) have spurred innovation in medical imaging solutions.
• Existing solutions facilitate medical practitioners in diagnosing lung cancer, but precise segmentation remains a challenge.

Purpose of the Study:

• To present an end-to-end neural network model for precise lung nodule segmentation.
• To develop a model suitable for cloud computing environments to aid in lung cancer diagnosis.
• To enhance the accuracy of lung nodule segmentation maps.

Main Methods:

• Developed a novel "semi-residual Multi-resolution Convolutional Neural Network" (semi-residual MCNN).
• Incorporated semi-residual building blocks, group normalization, and multi-resolution output heads into the network architecture.
• Trained and tested the model using the LIDC-IDRI dataset, comprising 1,018 lung CT images.

Main Results:

• The semi-residual MCNN demonstrated enhanced predictive accuracy for lung nodule segmentation.
• The model's architecture, featuring specific components, contributed to improved segmentation precision.
• Rigorous testing on the LIDC-IDRI dataset validated the model's performance.

Conclusions:

• The proposed semi-residual MCNN offers a precise solution for lung nodule segmentation in cloud environments.
• The model's innovative architectural features contribute to improved accuracy in medical image analysis.
• This approach has the potential to significantly aid medical practitioners in lung cancer diagnosis.