Training a neural network with images captured by dual-energy CT can produce high quality studies without the added dose or expense.
Adding a deep learning algorithm to standard CT technology can produce images that offer greater detail without the added radiation, new research has revealed.
In a study published Oct. 19 in Patterns, a team of engineers from Rensselaer Polytechnic Institute demonstrated how using this type of algorithm with conventional CT scans can capture images that typically require dual-energy CT (DECT).
“We hope that this technique will help extract more information from a regular single-spectrum X-ray CT scan, make it more quantitative, and improve diagnosis,” said study co-author Ge Wang, an endowed professor of biomedical engineering and director of the Biomedical Imaging Center in the Center for Biotechnology and Interdisciplinary Studies (CBIS).
Although CT scans are widely used in the United States – studies have spiked nearly 30-fold in the past 40 years to more than 80 million annual exams – they cannot give providers enough detail about the composition of imaged tissues. DECT can offer that information, but it comes with both higher radiation dose and cost.
Related Content: Dual-Energy CT in the ED Brings Greater Confidence
To avoid these issues and still give radiologists the information necessary for more informed interpretations, Wang’s team trained their neural network, dubbed ResNet DL, on DECT-derived virtual monoenergetic (VM) images. Their deep learning approach is designed for extraction of a non-linear transform from a single-spectrum CT image training dataset to reconstruct VM images using only a single-spectrum dataset.
According to their tests, this trained neural network produced high quality approximations of VM images with a relative error rate of less than 2 percent, Wang said. The structural similarity was 0.9991 +/- 0.0018, and it showed structural information, particularly texture features, were well preserved by the machine learning method. In addition, bone images were clearly separated from reconstructed VM images.
“It has been clearly demonstrated in our study that a conventional CT energy-integrating dataset coupled with [deep learning] can deliver a close approximation of DECT images,” the team said. “Thus, it is potentially feasible to just use conventional CT to perform some important tasks of DECT.”
Ultimately, from a clinical perspective, they said, the potential exists for this VM imaging method to significantly reduce scan times and radiation dose associated with a photon-counting micro-CT scan.
“Professor Wang and his team’s expertise in bioimaging is giving physicians and surgeons ‘new eyes’ in diagnosing and treating disease,” said Deepak Vashishth, CBIS director. “This research effort is a prime example of the partnership needed to personalize and solve persistent human health challenges.”
For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.