Coded-Aperture Compressed Sensing for Image Guidance in Radiation Therapy

Abstract

The ability of X-ray Fluorescence (XF) imaging to deliver molecular information in addition to standard IGRT is powerful and promises advancements in quality and efficiency of cancer treatment. While first experiments in XF Computed Tomography (XFCT) achieved sufficient molecular sensitivity for clinical applications, scanning times in the order of hours reduce its applicability. To circumvent this restriction, a Coded-Aperture Compressed sensing (CAC) approach is proposed to boost the data acquisition times considerably while keeping the loss of molecular sensitivity at a minimum. This approach is promising for the establishment of Molecular Guided Radiation Therapy (MGRT) in the clinic. To study the performance of CAC-XFCT imaging, a numerical simulation was established to emulate the beam encoding, the contrast agent excitation, the fluorescence emission, and event acquisition. A cylindrical head phantom with five target lesions, filled with cisplatin, was imaged with different excitation modes. The phantom was irradiated with different sets of pencil beams on different rotational and rotational positions around the phantom. For the coded aperture implementation, the beam sets consisting of 2, 4, 8, 16, 32, and 64 simultaneous beams were randomly chosen with a binary mask. The compressed sensing relies on the sparsity of the contrast signal and the displacement from the coding mask. The system matrix is calculated based on the Siddon algorithm by integrating the line integrals of the contrast agent distribution along the beam directions. The system matrix is updated based on the excitation pattern of the binary mask. To guarantee sparsity, the system matrix is multiplied with a wavelet transform matrix. The inverse problem is finally solved with Tikhonov regularization and L1 norm minimization. The quality of reconstructed multi-beam CAC-XFCT images, measured as the normalized mean square error (NMSE), and contrast recovery coefficient (CRC), is barely affected by incorporating more pencil beams in the acquisition process. The NMSE is increased from 0.41 for single pencil beam to 0.54 and 0.56 for 8 and 16 simultaneous pencil beams, respectively. The CRC is reduced from 0.65 to 0.6 and 0.52, respectively. Following these results, one hour of single pencil beam XFCT acquisitions for clinical applications can be acquired in less than 5 minutes. It is presented that the principle of CAC-XFCT imaging leads to astonishing accelerations in acquisition times compared to standard XFCT. These results for the imaging of high atomic number probes in human sized objects is promising to find applications in the clinical context. Especially cisplatin imaging is interesting, e.g. for a combined radio-chemo therapy or to monitor the response of immuno-therapy markers during treatment. It is demonstrated that scanning times in the order of minutes are accessible which makes CAC-XFCT extreme alluring for clinically relevant applications.