Related Experiment Video
Updated: Oct 2, 2025
Enhancing Electrode Location Assessment in Cochlear Implantation via Computed Tomography Image Fusion
Published on: January 17, 2025
Multi-modal feature-fusion for CT metal artifact reduction using edge-enhanced generative adversarial networks
Zhiwei Huang1, Guo Zhang1, Jinzhao Lin2
1School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China; School of Medical Information and Engineering, Southwest Medical University, Luzhou, 646000, China; Chongqing Key Laboratory of Photo-electronic Information Sensing and Transmitting Technology, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China.
This study introduces a new generative adversarial network method to reduce metal artifacts in Computed Tomography (CT) imaging. The approach effectively reduces artifacts while enhancing image texture and structure for clearer diagnoses.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Computed Tomography (CT) is vital for organ screening and disease diagnosis.
- Metallic implants in patients cause significant metal artifacts in CT images, degrading image quality.
- Existing metal artifact reduction methods are often inadequate, leading to artifacts and poor image quality.
Purpose of the Study:
- To develop a novel metal artifact reduction method for CT images.
- To simultaneously reduce metal artifacts and enhance texture structure in corrected CT images.
- To improve the diagnostic quality of CT images affected by metallic implants.
Main Methods:
- A novel metal artifact reduction method based on generative adversarial networks (GANs).
- Incorporation of multi-modal feature fusion using interactive text and CT imaging data.
- Design of an edge-enhance sub-network to prevent secondary artifacts and suppress noise.
Main Results:
- Achieved an average increment of 11.3% in Peak Signal-to-Noise Ratio (PSNR) and 12.1% in Structural Similarity Index Measure (SSIM) on the DeepLesion dataset.
- Physician evaluations showed superior performance in sharpness (6.3%), resolution (7.1%), invariance (5.50%), and acceptability (6.9%) compared to existing methods.
- Demonstrated high-quality metal artifact reduction results with enhanced texture structure.
Conclusions:
- The proposed multi-modal GAN-based method effectively reduces metal artifacts in CT images.
- The method enhances image quality, ensuring symptom consistency and improving diagnostic accuracy.
- This approach offers a significant advancement in addressing the challenges of metal artifacts in medical imaging.
