TU-EF-204-05: Task-Driven Cerebral CT Perfusion Imaging

Purpose: This work introduces a task-driven framework to cerebral CT perfusion (CTP) imaging to improve the image quality and quantification accuracy of parametric perfusion maps used in the clinical assessment of acute ischemic strokes and stroke mimics. Methods: First, the concepts of cerebral anatomical noise, noise power spectrum (NPS), MTF, and generalized task-based detectability index were developed for CTP imaging. Second, a spatial-temporal (4D) cascaded system model was developed to establish an analytical and quantitative correspondence between system parameters and CTP imaging performance. Third, a library of CTP imaging tasks such as the detection of ischemic lesions with different sizes and severities of perfusion deficiencies were created under the guidance of a neurosurgeon. Finally, data acquisition, reconstruction, and processing parameters were prospectively optimized for each CTP imaging task for improved imaging performance. The proposed method was validated in both digital anthropomorphic 4D perfusion phantoms and in vivo canine stroke models. Results: The NPS, MTF, and the generalized task-based detectability index predicted by the 4D cascaded model matched exactly with the experimental results. For both non-deconvolution-based and deconvolution-based perfusion calculation methods, as long as the signal and noise properties of the input source CT images are known, the image quality and bias of the final CTP maps can be predicted. Guided by the tasked-driven framework, a novel tube current modulation technique for the acquisition of source images was developed, and parameters involved in parametric perfusion map calculations were optimized. Through these innovations, perfusion maps of the canines demonstrated 80% reduction in image noise as well as significant improvement in the detectability of ischemic lesions and the quantitative accuracy of perfusion measurements. Conclusion: The task-driven approach has successfully guided the improvement of conventional CTP imaging techniques, potentially enabling a leap in the image quality and quantification accuracy of parametric perfusion maps. K. Li, K. Niu, Y. Wu, P. Yang: Nothing to disclose. G.-H. Chen: Research funded, GE Healthcare; Research funded, Siemens AX.