Parallel processing model for low-dose computed tomography image denoising.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Low-dose computed tomography (LDCT) reduces patient radiation exposure but produces noisy images, hindering accurate diagnosis.
• Current deep learning (DL) denoising methods struggle with variable data from different imaging sources.
• A need exists for robust DL models that can generalize across diverse LDCT data.

Purpose of the Study:

• To develop a novel deep learning model for effective denoising of LDCT images from multiple sources.
• To address the limitations of existing DL denoising techniques in handling heterogeneous LDCT data.
• To enhance the diagnostic quality of LDCT images through improved noise reduction and structure preservation.

Main Methods:

• Proposed the Multi-Encoder Deep Feature Transformation Network (MDFTN), a parallel processing model.
• MDFTN utilizes multiple encoders for parallel feature extraction and a Deep Feature Transformation Module (DFTM) for feature compression into a shared space.
• Collaborative training of encoders and decoders enables simultaneous processing of multisource LDCT data.

Main Results:

• The MDFTN successfully processed multisource LDCT data within a unified framework.
• Demonstrated significant noise suppression and preservation of fine structures in LDCT images.
• Experiments on public and local datasets validated the model's adaptability and generalization capabilities.

Conclusions:

• The MDFTN offers an effective solution for denoising LDCT images from diverse sources.
• The parallel processing and feature transformation approach enhances model performance and generalizability.
• This method has the potential to improve diagnostic accuracy in clinical LDCT applications.