Mask-aware transformer with structure invariant loss for CT translation

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Multi-phase enhanced computed tomography (MPECT) is crucial for liver lesion detection but carries allergy risks.
• Current CT translation methods overlook valuable clinical domain knowledge, limiting their effectiveness.

Purpose of the Study:

• To develop an automatic CT translation method that incorporates clinical domain knowledge for improved MPECT image generation.
• To enhance liver lesion detection and reduce patient allergy risks associated with MPECT examinations.

Main Methods:

• Proposed a Mask-Aware Transformer (MAFormer) model for CT translation.
• Introduced a mask estimator to predict subtraction images, highlighting contrast-enhanced regions.
• Developed Mask-Aware Transformer based Normalization (MATNorm) for integrating subtraction images and capturing dependencies.
• Implemented a structure invariant loss to preserve biological structures in translated CT images.

Main Results:

• The MAFormer model effectively translates plain CT to MPECT images.
• The method demonstrates superiority over existing state-of-the-art CT translation techniques.
• Experimental results validate the effectiveness of incorporating clinical domain knowledge.

Conclusions:

• The MAFormer represents a significant advancement in automatic CT translation for medical imaging.
• This approach holds promise for improving diagnostic accuracy and patient safety in liver disease diagnosis.