Combined multi-kernel head computed tomography images optimized for depicting both brain parenchyma and bone

Continuous Conversion of CT Kernel Using Switchable CycleGAN With AdaIN

X-ray computed tomography (CT) uses different filter kernels to highlight different structures. Since the raw sinogram data is usually removed after the reconstruction, in case there is additional need for other types of kernel images that were not previously generated, the patient may need to be scanned again. Accordingly, there exists increasing demand for post-hoc image domain conversion from one kernel to another without sacrificing the image quality. In this paper, we propose a novel unsupervised continuous kernel conversion method using cycle-consistent generative adversarial network (cycleGAN) with adaptive instance normalization (AdaIN). Even without paired training data, not only can our network translate the images between two different kernels, but it can also convert images along the interpolation path between the two kernel domains. We also show that the quality of generated images can be further improved if intermediate kernel domain images are available. Experimental results confirm that our method not only enables accurate kernel conversion that is comparable to supervised learning methods, but also generates intermediate kernel images in the unseen domain that are useful for hypopharyngeal cancer diagnosis.

Image Quality Required for the Diagnosis of Skull Fractures Using Head CT: A Comparison of Conventional and Improved Reconstruction Kernels

BACKGROUND AND PURPOSE

Although skull fractures are generally assessed on bone images obtained by using head CT, the combined multikernel technique that enables evaluation of both brain and bone through a change in the window settings of an image set has been reported. The purpose of this retrospective study was to determine the image quality required for the accurate assessment of skull fractures by using head CT.

MATERIALS AND METHODS

A random sample of 50 patients (25 nonfracture and 25 simple nondisplaced skull fractures) was selected, and sets of conventional brain and bone images and improved combined multikernel images were reconstructed (4614 images). Three radiologists indicated their confidence levels regarding the presence of skull fractures by marking on a continuous scale for each image set. The mean area under the receiver operating characteristic curve was calculated for each kernel, and the statistical significance of differences was tested by using the Dorfman-Berbaum-Metz method.

RESULTS

Although a difference in the diagnostic performance of the 3 radiologists was suggested, the mean area under the curve value showed no significant differences among the 3 reconstruction kernels (P = .95 [bone versus combined]), P = .91 [bone versus brain]), and P = .88 [brain versus combined]). However, the quality of brain images was distinctly poorer than the quality of the other 2 images.

CONCLUSIONS

There was no significant difference in the diagnostic performance of brain, bone, and combined multikernel images for skull fractures. Skull fracture diagnosis is made possible by brain image assessments. Combined multikernel images offer the advantage of high-quality brain and bone images.