Improving image quality of sparse-view lung tumor CT images with U-Net.

BACKGROUND: We aimed to improve the image quality (IQ) of sparse-view computed tomography (CT) images using a U-Net for lung metastasis detection and determine the best tradeoff between number of views, IQ, and diagnostic confidence. METHODS: CT images from 41 subjects aged 62.8 ± 10.6 years (mean ± standard deviation, 23 men), 34 with lung metastasis, 7 healthy, were retrospectively selected (2016-2018) and forward projected onto 2,048-view sinograms. Six corresponding sparse-view CT data subsets at varying levels of undersampling were reconstructed from sinograms using filtered backprojection with 16, 32, 64, 128, 256, and 512 views. A dual-frame U-Net was trained and evaluated for each subsampling level on 8,658 images from 22 diseased subjects. A representative image per scan was selected from 19 subjects (12 diseased, 7 healthy) for a single-blinded multireader study. These slices, for all levels of subsampling, with and without U-Net postprocessing, were presented to three readers. IQ and diagnostic confidence were ranked using predefined scales. Subjective nodule segmentation was evaluated using sensitivity and Dice similarity coefficient (DSC); clustered Wilcoxon signed-rank test was used. RESULTS: The 64-projection sparse-view images resulted in 0.89 sensitivity and 0.81 DSC, while their counterparts, postprocessed with the U-Net, had improved metrics (0.94 sensitivity and 0.85 DSC) (p = 0.400). Fewer views led to insufficient IQ for diagnosis. For increased views, no substantial discrepancies were noted between sparse-view and postprocessed images. CONCLUSIONS: Projection views can be reduced from 2,048 to 64 while maintaining IQ and the confidence of the radiologists on a satisfactory level. RELEVANCE STATEMENT: Our reader study demonstrates the benefit of U-Net postprocessing for regular CT screenings of patients with lung metastasis to increase the IQ and diagnostic confidence while reducing the dose. KEY POINTS: • Sparse-projection-view streak artifacts reduce the quality and usability of sparse-view CT images. • U-Net-based postprocessing removes sparse-view artifacts while maintaining diagnostically accurate IQ. • Postprocessed sparse-view CTs drastically increase radiologists' confidence in diagnosing lung metastasis.

Improving Automated Hemorrhage Detection at Sparse-View CT via U-Net-based Artifact Reduction.

Purpose To explore the potential benefits of deep learning-based artifact reduction in sparse-view cranial CT scans and its impact on automated hemorrhage detection. Materials and Methods In this retrospective study, a U-Net was trained for artifact reduction on simulated sparse-view cranial CT scans in 3000 patients, obtained from a public dataset and reconstructed with varying sparse-view levels. Additionally, EfficientNet-B2 was trained on full-view CT data from 17 545 patients for automated hemorrhage detection. Detection performance was evaluated using the area under the receiver operating characteristic curve (AUC), with differences assessed using the DeLong test, along with confusion matrices. A total variation (TV) postprocessing approach, commonly applied to sparse-view CT, served as the basis for comparison. A Bonferroni-corrected significance level of .001/6 = .00017 was used to accommodate for multiple hypotheses testing. Results Images with U-Net postprocessing were better than unprocessed and TV-processed images with respect to image quality and automated hemorrhage detection. With U-Net postprocessing, the number of views could be reduced from 4096 (AUC: 0.97 [95% CI: 0.97, 0.98]) to 512 (0.97 [95% CI: 0.97, 0.98], P < .00017) and to 256 views (0.97 [95% CI: 0.96, 0.97], P < .00017) with a minimal decrease in hemorrhage detection performance. This was accompanied by mean structural similarity index measure increases of 0.0210 (95% CI: 0.0210, 0.0211) and 0.0560 (95% CI: 0.0559, 0.0560) relative to unprocessed images. Conclusion U-Net-based artifact reduction substantially enhanced automated hemorrhage detection in sparse-view cranial CT scans. Keywords: CT, Head/Neck, Hemorrhage, Diagnosis, Supervised Learning Supplemental material is available for this article. © RSNA, 2024.

Feasibility of Dark-Field Radiography to Enhance Detection of Nondisplaced Fractures.

Background Many clinically relevant fractures are occult on conventional radiographs and therefore challenging to diagnose reliably. X-ray dark-field radiography is a developing method that uses x-ray scattering as an additional signal source. Purpose To investigate whether x-ray dark-field radiography enhances the depiction of radiographically occult fractures in an experimental model compared with attenuation-based radiography alone and whether the directional dependence of dark-field signal impacts observer ratings. Materials and Methods Four porcine loin ribs had nondisplaced fractures experimentally introduced. Microstructural changes were visually verified using high-spatial-resolution three-dimensional micro-CT. X-ray dark-field radiographs were obtained before and after fracture, with the before-fracture scans serving as control images. The presence of a fracture was scored by three observers using a six-point scale (6, surely; 5, very likely; 4, likely; 3, unlikely; 2, very unlikely; and 1, certainly not). Differences between scores based on attenuation radiographs alone (n = 96) and based on combined attenuation and dark-field radiographs (n = 96) were evaluated by using the DeLong method to compare areas under the receiver operating characteristic curve. The impact of the dark-field signal directional sensitivity on observer ratings was evaluated using the Wilcoxon test. The dark-field data were split into four groups (24 images per group) according to their sensitivity orientation and tested against each other. Musculoskeletal dark-field radiography was further demonstrated on human finger and foot specimens. Results The addition of dark-field radiographs was found to increase the area under the receiver operating characteristic curve to 1 compared with an area under the receiver operating characteristic curve of 0.87 (95% CI: 0.80, 0.94) using attenuation-based radiographs alone (P < .001). There were similar observer ratings for the four different dark-field sensitivity orientations (P = .16-.65 between the groups). Conclusion These results suggested that the inclusion of dark-field radiography has the potential to help enhance the detection of nondisplaced fractures compared with attenuation-based radiography alone. © RSNA, 2024 See also the editorial by Rubin in this issue.

The transcription factor prospero homeobox protein 1 is a direct target of SoxC proteins during developmental vertebrate neurogenesis.

The high-mobility-group domain containing SoxC transcription factors Sox4 and Sox11 are expressed and required in the vertebrate central nervous system in neuronal precursors and neuroblasts. To identify genes that are widely regulated by SoxC proteins during vertebrate neurogenesis we generated expression profiles from developing mouse brain and chicken neural tube with reduced SoxC expression and found the transcription factor prospero homeobox protein 1 (Prox1) strongly down-regulated under both conditions. This led us to hypothesize that Prox1 expression depends on SoxC proteins in the developing central nervous system of mouse and chicken. By combining luciferase reporter assays and over-expression in the chicken neural tube with in vivo and in vitro binding studies, we identify the Prox1 gene promoter and two upstream enhancers at -44 kb and -40 kb relative to the transcription start as regulatory regions that are bound and activated by SoxC proteins. This argues that Prox1 is a direct target gene of SoxC proteins during neurogenesis. Electroporations in the chicken neural tube furthermore show that Prox1 activates a subset of SoxC target genes, whereas it has no effects on others. We propose that the transcriptional control of Prox1 by SoxC proteins may ensure coupling of two types of transcription factors that are both required during early neurogenesis, but have at least in part distinct functions. Open Data: Materials are available on https://cos.io/our-services/open-science-badges/ https://osf.io/93n6m/.

The closely related transcription factors Sox4 and Sox11 function as survival factors during spinal cord development.

Development of the mouse CNS was reported to be normal in the absence of either Sox4 or its close relative Sox11 despite strong and widespread expression of both transcription factors. In this study, we show that combined absence of both Sox proteins in the mouse leads to severe hypoplasia of the developing spinal cord. Proliferation of neuroepithelial precursor cells in the ventricular zone was unaffected. These cells also acquired their correct positional identity. Both glial and neuronal progenitors were generated and neurons appeared in a similar spatiotemporal pattern as in the wild-type. Rates of cell death were however dramatically increased throughout embryogenesis in the double deficient spinal cord arguing that Sox4 and Sox11 are jointly and redundantly required for cell survival. The absence of pronounced proliferation, patterning, specification, and maturation defects furthermore indicates that the decreased cell survival is not a secondary effect of one of these events. We therefore conclude that the two Sox proteins directly function as pro-survival factors during spinal cord development in neural cell types.

Ectopic T-cell specificity and absence of perforin and granzyme B alleviate neural damage in oligodendrocyte mutant mice.

In transgenic mice overexpressing the major myelin protein of the central nervous system, proteolipid protein, CD8+ T-lymphocytes mediate the primarily genetically caused myelin and axon damage. In the present study, we investigated the cellular and molecular mechanisms underlying this immune-related neural injury. At first, we investigated whether T-cell receptors (TCRs) are involved in these processes. For this purpose, we transferred bone marrow from mutants carrying TCRs with an ectopic specificity to ovalbumin into myelin mutant mice that also lacked normal intrinsic T-cells. T-lymphocytes with ovalbumin-specific TCRs entered the mutant central nervous system to a similar extent as T-lymphocytes from wild-type mice. However, as revealed by histology, electron microscopy and axon- and myelin-related immunocytochemistry, these T-cells did not cause neural damage in the myelin mutants, reflecting the need for specific antigen recognition by cytotoxic CD8+ T-cells. By chimerization with bone marrow from perforin- or granzyme B (Gzmb)-deficient mice, we demonstrated that absence of these cytotoxic molecules resulted in reduced neural damage in myelin mutant mice. Our study strongly suggests that pathogenetically relevant immune reactions in proteolipid protein-overexpressing mice are TCR-dependent and mediated by the classical components of CD8+ T-cell cytotoxicity, perforin, and Gzmb. These findings have high relevance with regard to our understanding of the pathogenesis of disorders primarily caused by genetically mediated oligodendropathy.