Dual-energy imaging in stroke : feasibility of dual-layer detector cone-beam computed tomography

Abstract: Background: Dual-energy computed tomography (DECT) is increasingly available and used in the standard diagnostic setting of ischemic stroke patients. For stroke patients with suspected large vessel occlusion, cone-beam computed tomography (CBCT) in the interventional suite could be an alternative to CT to shorten door to thrombectomy time. This approach could potentially lead to an improved patient outcome. However, image quality in CBCT is typically limited by artifacts and poor differentiation between gray and white matter. A dual-layer detector CBCT (DL-CBCT) system could be used to separate photon energy spectra with the potential to increase visibility of clinically relevant features, and acquire additional information. Purpose: Paper I evaluated how a range of DECT virtual monoenergetic images (VMI) impact identification of early ischemic changes, compared to conventional polyenergetic CT images. Paper II characterized the performance of a novel DLCBCT system with regards to clinically relevant imaging features. Paper III & IV investigated if DL-CBCT VMIs are sufficient for stroke diagnosis in the interventional suite, compared to reference standard CT. Methods: Paper I was a retrospective single-center study including consecutive patients presenting with acute ischemic stroke caused by an occlusion of the intracranial internal carotid artery or proximal middle cerebral artery. Automated Alberta Stroke Program Early Computed Tomography Score (ASPECTS) results from conventional images and 40-120 keV VMI were generated and compared to reference standard CT ASPECTS. In paper II, a prototype dual-layer detector was fitted into a commercial interventional C-arm CBCT system to enable dual-energy acquisitions. Metrics for spatial resolution, noise and uniformity were gathered. Clinically relevant tissue and iodine substitutes were characterized in terms of effective atomic numbers and electron densities. Iodine quantification was performed and virtual non-contrast (VNC) images were evaluated. VMIs were reconstructed and used for CT number estimation and evaluation of contrast-to-noise ratios (CNR) in relevant tissue pairings. In paper III and IV, a prospective single-center study enrolled consecutive participants with ischemic or hemorrhagic stroke on CT. In paper III, hemorrhage detection accuracy, ASPECTS accuracy, subjective and objective image quality were evaluated on non-contrast DL-CBCT 75 keV VMI and compared to reference standard CT. In paper IV, intracranial arterial segment vessel visibility and artifacts were evaluated on intravenous DL-CBCT angiography (DL-CBCTA) 70 keV VMI and compared to CT angiography (CTA). In both paper III and IV, non-inferiority was determined by the exact binomial test with a one-sided lower performance boundary set to 80% (98.75% CI). Main results: In paper I, 24 patients were included. 70 keV VMI had the highest region-based ASPECTS accuracy (0.90), sensitivity (0.82) and negative predictive value (0.94), whereas 40 keV VMI had the lowest accuracy (0.77), sensitivity (0.34) and negative predictive value (0.80). In paper II, the prototype and commercial CBCT had a similar spatial resolution and noise using the same standard reconstruction. For all tissue substitutes, the mean accuracy in effective atomic number was 98.2% (SD 1.2%) and 100.3% (SD 0.9%) for electron density. Iodine quantification had a mean difference of -0.1 (SD 0.5) mg/ml compared to the true concentrations. For VNC images, iodine substitutes with blood averaged 43.2 HU, blood only 44.8 HU, iodine substitutes with water 2.6 HU. A noise-suppressed dataset showed a CNR peak at 40 keV VMI and low at 120 keV VMI. In the same dataset without noise suppression, peak CNR was seen at 70 keV VMI and a low at 120 keV VMI. CT numbers of various clinically relevant objects generally matched the calculated CT number in a wide range of VMIs. In paper III, 27 participants were included. One reader missed a small bleeding, however all hemorrhages were detected in the majority analysis (100% accuracy, CI lower boundary 86%, p=0.002). ASPECTS majority analysis had 90% accuracy (CI lower boundary 85%, p<0.001), sensitivity was 66% (individual readers 67%, 69% and 76%), specificity was 97% (97%, 96% and 89%). Subjective and objective image quality metrics were inferior to CT. In paper IV, 21 participants had matched image sets. After excluding examinations with scan issues, all readers considered DL-CBCTA non-inferior to CTA (CI boundary 93%, 84%, 80%, respectively), when assessing arteries relevant in candidates for intracranial thrombectomy. Artifacts were more prevalent compared to CTA. Conclusions: In paper I, automated 70 keV VMI ASPECTS had the highest diagnostic accuracy, sensitivity and negative predictive value overall. Different VMI energy levels impact the identification of early ischemic changes on DECT. In paper II, the DL-CBCT prototype system showed comparable technical metrics to a commercial CBCT system, while offering dual-energy capability. The dual-energy images indicated a consistent ability to separate and characterize clinically relevant tissues, blood and iodine. Thus, the DL-CBCT system could find utility in the diagnostic setting. In paper III, non-contrast DL-CBCT 75 keV VMI showed non-inferior hemorrhage detection and ASPECTS accuracy to CT. However, image quality was inferior compared to CT, and visualization of small subarachnoid hemorrhages after treatment remains a challenge. In the same stroke cohort, paper IV showed non-inferior vessel visibility for DL-CBCTA 70 keV VMI compared to CTA under certain conditions. Specifically, the prototype system had a long scan time and was not capable of bolus tracking which resulted in scan issues. After excluding participants with such issues, DL-CBCTA 70 keV VMI were found non-inferior to CTA. In summary, the findings of this thesis indicate that DL-CBCT may be sufficient for stroke assessment in the interventional suite with the potential to bypass CT in patients eligible for thrombectomy. However, issues related to the prototype system and the visualization of small hemorrhages highlight the need of further development.