2023 World Conference on Lung Cancer (Posters)-EP05.01. Multi-scale Generator with Channel-wise Mask Attention to Generate Synthetic Contrast-enhanced Chest Computed Tomography

EP05.01. Multi-scale Generator with Channel-wise Mask Attention to Generate Synthetic Contrast-enhanced Chest Computed Tomography - PDF(Slides)

Back to course

Pdf Summary

Researchers have proposed a multi-scale generator with a channel-wise mask attention module (MGCMA) to generate synthetic contrast-enhanced chest computed tomography (CECT) images from non-contrast CT (NCCT) images. The administration of contrast media in CECT can have adverse effects, but it provides valuable information for diagnosing internal pathologies. The proposed generator architecture considers multiple scales to learn structural and texture information. It is trained using a combination of real CECT and NCCT images. The performance of MGCMA was compared with other image-to-image translation methods using similarity metrics, and it consistently outperformed in terms of peak signal-to-noise ratio (PSNR) values. The qualitative assessments showed that MGCMA preserved organ shape and enhanced selective regions where blood flow occurs. 75% of the doctors found the generated images acceptable without changes, and the remaining 70% found them generally acceptable with minor changes. The researchers concluded that their model outperformed other baseline models in various metrics and achieved plausible outcomes according to clinicians' evaluations. In the future, they plan to apply their approach to patient images with thoracic malignancies.

Asset Subtitle

Yun-Gyoo Lee

Meta Tag

Speaker Yun-Gyoo Lee

Topic Pulmonology & Staging

Keywords

multi-scale generator

channel-wise mask attention module

synthetic contrast-enhanced chest computed tomography

non-contrast CT

contrast media

diagnosing internal pathologies

peak signal-to-noise ratio

organ shape preservation

blood flow enhancement

thoracic malignancies