Low-dose computed tomography image reconstruction via a multistage convolutional neural network with autoencoder...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Computed tomography (CT) is vital for medical diagnosis but carries risks from radiation exposure.
• Low-dose CT (LDCT) reduces radiation but introduces noise and artifacts, compromising image quality.
• Effective reconstruction methods are crucial for maintaining diagnostic accuracy with LDCT.

Purpose of the Study:

• To develop an advanced image reconstruction method for low-dose CT (LDCT) scans.
• To address the challenges of noise and artifact reduction in LDCT images.
• To improve the diagnostic capability of LDCT by enhancing image quality.

Main Methods:

• A multistage convolutional neural network (MSCNN) framework was proposed for LDCT image reconstruction.
• A dilated residual convolutional neural network (DRCNN) was employed for image denoising.
• A novel self-calibration module (SCM) and an autoencoder perceptual loss network were integrated to refine features and preserve details.

Main Results:

• The proposed network achieved high quantitative scores for peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and visual information fidelity (VIF).
• Qualitative analysis demonstrated a superior balance between noise elimination and preservation of image details.
• Ablation studies confirmed the effectiveness of individual network modules and the loss function.

Conclusions:

• A novel LDCT image reconstruction method combining autoencoder perceptual loss networks with MSCNN was successfully developed.
• The proposed method significantly outperforms existing techniques in both quantitative metrics and visual evaluation.
• This approach enhances the diagnostic utility of low-dose CT imaging.