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(Abstract)

Back to course

Pdf Summary

This document presents a study that aims to generate synthetic contrast-enhanced chest computed tomography (CECT) images from non-contrast CT (NCCT) images. The authors propose a deep learning model called MGCMA (multi-scale generator with channel-wise mask attention) for this purpose. The study used a dataset of NCCT and CECT images from 207 participants, split into a development set and a test set. The proposed generator architecture utilizes a multi-scale feature pyramid network and incorporates a channel-wise mask attention module. The performance of MGCMA was compared to other image-to-image translation methods using peak signal-to-noise ratio (PSNR) values. The results showed that MGCMA consistently outperformed the other methods in terms of PSNR values in all test sets. In addition, qualitative assessments conducted by physicians showed that the generated images from MGCMA were generally acceptable and preserved organ shape. The study concludes that MGCMA is a promising model for generating synthetic CECT images and has the potential to be applied to patient images with thoracic malignancies in the future. The keywords associated with this study are synthetic contrast, chest computed tomography, and the track of the session is pulmonology and staging.

Asset Subtitle

Yun-Gyoo Lee

Meta Tag

Speaker Yun-Gyoo Lee

Topic Pulmonology & Staging

Keywords

synthetic contrast

chest computed tomography

deep learning model

MGCMA

non-contrast CT

NCCT

contrast-enhanced CT

image generation

peak signal-to-noise ratio

organ shape preservation