Development of A deep Learning-based algorithm for High-Pitch helical computed tomography imaging
Xiaoman Duan,Xiao Fan Ding,Samira Khoz,Xiongbiao Chen,Ning Zhu
DOI: https://doi.org/10.1016/j.eswa.2024.125663
IF: 8.5
2024-11-07
Expert Systems with Applications
Abstract:High-pitch X-ray helical computed tomography (HCT) imaging has been recently drawing considerable attention in biomedical fields due to its ability to reduce the scanning time and thus lower the radiation dose that objects (being imagined) may receive. However, the issue of compromised reconstruction quality caused by incomplete data in these high-pitch CT scans remains, thus limiting its applications. By addressing the aforementioned issue, this paper presents our study on the development of a novel deep leaning (DL)-based algorithm, ViT-U, for high-pitch X-ray propagation-based imaging HCT (PBI-HCT) reconstruction. ViT-U consists of two key process modules of a vision transformer (ViT) and a convolutional neural network (i.e., U-Net), where ViT addresses the missing information in the data domain and U-Net enhances the post data-processing in the reconstruction domain. For verification, we designed and conducted simulations and experiments with both low-density-biomaterial samples and biological-tissue samples to exemplify the biomedical applications, and then examined the ViT-U performance with varying pitches of 3, 3.5, 4, and 4.5, respectively, for comparison in term of radiation does and reconstruction quality. Our results showed that the high-pitch PBI-HCT allowed for the dose reduction from 72% to 93%. Importantly, our results demonstrated that the ViT-U exhibited outstanding performance by effectively removing the missing wedge artifacts thus enhancing the reconstruction quality of high-pitch PBI-HCT imaging. Also, our results showed the superior capability of ViT-U to achieve high quality of reconstruction from the high-pitch images with the helical pitch value up to 4 (which allowed for the substantial reduction of radiation doses). Taken together, our DL-based ViT-U algorithm not only enables high-speed imaging with low radiation dose, but also maintains the high quality of imaging reconstruction, thereby offering significant potentials for biomedical imaging applications.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science