In the United States, more than 80 million CT scans are performed each year. A CT scan (computer tomography) is a common and effective way for doctors to look inside the human body to diagnose, treat, and monitor disease. However, a scanner does not tell the whole story. Although it reveals the morphology of tissues, it does not provide any information on the elemental composition of tissues. A workaround is to use contrast agents like iodine, but even this method has pitfalls – contrast-enhanced structures may be similar in density to bone or calcified plaques, making them difficult to distinguish .
So doctors turned to dual-energy computed tomography, which brings together two sets of data to produce images that reveal both tissue shape and tissue composition information. But dual-energy CT scans are expensive, high-tech, and often require a higher radiation dose than conventional CT scans.
So engineers at Rensselaer Polytechnic Institute turned to deep learning to help them.
“With traditional CT, you take a grayscale image, but with dual-energy CT, you take a two-color image,” explained Ge Wang, professor of biomedical engineering at Rensselaer Polytechnic Institute and author. of a new article on the subject. “With deep learning, we’re trying to use the standard machine to do the dual-energy CT imaging job.”
Using virtual monoenergetic (VM) images derived from dual-energy CT, Wang and his team trained the ResNet deep learning network to map single-spectrum CT images to virtual monoenergetic images at d predefined energy.
According to their findings, described in the review Grounds, the trained neural network provided high-quality approximations of dual-energy CT-derived VM images with a relative error of less than 2% for the test dataset. Moreover, the structural information, especially the texture features, was well preserved by the machine learning method.
The researchers then used the learned VM images to generate multi-material decomposition (MMD) images, hoping to obtain close approximations to the clinical images directly produced by the dual energy. Again, the results demonstrated high-quality material-specific images. The study points to a specific example in which a bone image was clearly separated from the reconstructed VM image, highlighting abdominal aortic calcification that was not clearly visible on conventional CT scans. Ultimately, the method enables multi-material decomposition into three tissue classes, with an accuracy comparable to dual-energy computed tomography.
“We hope this technique will help extract more information from a single-spectrum X-ray scanner, make it more quantitative, and improve diagnosis,” Wang said.
The researchers say their deep learning method is well suited for calculating proton stopping power for proton therapy planning, as well as photon-counting micro-CT for live preclinical applications. For preclinical imaging, the neural network could dramatically reduce scanning time and radiation dose by learning from multiple photo-counting micro-CT datasets and then reconstructing them to produce dual-energy CT images. .