Dual Energy CT is a modern imaging technique that is utilized in clinical practice to acquire spectral information for various diagnostic purposes including the identification, classification, and characterization of different liver lesions. It provides additional information that, when compared to the information available from conventional CT datasets, has the potential to benefit existing computer vision techniques by improving their accuracy and reliability. In order to evaluate the additional value of spectral versus conventional datasets when being used as input for machine learning algorithms, we implemented a weakly-supervised Convolutional Neural Network (CNN) that learns liver lesion localization and classification without pixel-level ground truth annotations. We evaluated the lesion classification (healthy, cyst, hypodense metastasis) and localization performance of the network for various conventional and spectral input datasets obtained from the same CT scan. The best results for lesion localization were found for the spectral datasets with distances of 8.22 ± 10.72 mm, 8.78 ± 15.21 mm and 8.29 ± 12.97 mm for iodine maps, 40 keV and 70 keV virtual mono-energetic images, respectively, while lesion localization distances of 10.58 ± 17.65 mm were measured for the conventional dataset. In addition, the 40 keV virtual mono-energetic datasets achieved the highest overall lesion classification accuracy of 0.899 compared to 0.854 measured for the conventional datasets. The enhanced localization and classification results that we observed for spectral CT data demonstrates that combining machine-learning technology with spectral CT information may improve the clinical workflow as well as the diagnostic accuracy.
Computed Tomography (CT) is one of the most important imaging modalities in the medical domain. Ongoing demand for reduction of the X-ray radiation dose and advanced reconstruction algorithms induce ultra-low dose CT acquisitions more and more. However, though advanced reconstructions lead to improved image quality, the ratio between electronic detector noise and incoming signal decreases in ultra-low dose scans causing a degradation of the image quality and, therefore, building a boundary for radiation dose reduction. Future generations of CT scanners may allow sparse sampled data acquisitions, where the source can be switched on and off at any source position. Sparse sampled CT acquisitions could reduce photon starvation in ultra-low dose scans by distributing the energy of skipped projections to the remaining ones. In this work, we simulated sparse sampled CT acquisitions from clinical projection raw data and evaluated the diagnostic value of the reconstructions compared to conventional CT. Therefore, we simulated radiation dose reduction with different degrees of sparse sampling and with a tube current simulator. Up to four experienced radiologists rated the diagnostic quality of each dataset. By a dose reduction to 25% of the clinical dosage, images generated with 4-times sparse sampling – meaning a gap of three projections between two sampling positions – were consistently rated as diagnostic, while about 20% of the ratings for conventional CT were non-diagnostic. Therefore, our data give an initial indication that with sparse sampling a reduction to 25% of the clinical dose is feasible without loss of diagnostic value.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.