Multi-domain CT translation by a routable translation network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Image Processing

Background:

• Computed tomography (CT) images exhibit significant variations due to differing scan parameters, reconstruction algorithms, and hardware designs across various manufacturers and sites.
• These variations pose a substantial challenge for quantitative analysis in multi-site or longitudinal studies, hindering consistent interpretation and comparison of imaging data.

Purpose of the Study:

• To develop a novel multi-domain image translation network for unifying the style of CT images from diverse sources.
• To enable consistent quantitative analysis of multi-domain CT images by minimizing variations introduced by imaging parameters and manufacturers.

Main Methods:

• Proposed a novel multi-domain image translation network featuring a shared encoder and a routable decoder architecture.
• The network utilizes a routing vector to control the translation process, allowing conversion of CT images from different scan parameters and manufacturers.
• Emphasized maximizing network expressivity and conditioning power through the proposed architecture.

Main Results:

• Experimental results demonstrated that the proposed CT image conversion effectively minimizes image characteristic variations caused by imaging parameters, reconstruction algorithms, and hardware designs.
• Quantitative evaluations and clinical assessments by radiologists confirmed the accuracy and effectiveness of the translation results.
• The method successfully addressed variations in CT images from multi-site or longitudinal studies.

Conclusions:

• The proposed multi-domain image translation network offers a robust solution for standardizing CT image styles across different sources.
• This technique is crucial for overcoming the challenges in quantitative evaluation of multi-domain CT images, thereby facilitating more reliable multi-site and longitudinal studies.
• The method holds significant potential for advancing quantitative analysis in medical imaging research and clinical practice.