dorsal/arxiv
View SchemaFIRST: Fast Iterative Reconstruction Software for (PET) tomography
| Authors | J. L. Herraiz, S. Espana, J. J. Vaquero, M. Desco, J. M. Udias |
|---|---|
| Categories | |
| ArXiv ID | physics/0609104 |
| URL | https://arxiv.org/abs/physics/0609104 |
| DOI | 10.1088/0031-9155/51/18/007 |
| Journal | Physics in Medicine and Biology, Volume 51, Number 18, 21 September 2006 |
Abstract
Small animal PET scanners require high spatial resolution and good sensitivity. To reconstruct high-resolution images in 3D-PET, iterative methods, such as OSEM, are superior to analytical reconstruction algorithms, although their high computational cost is still a serious drawback. The higher performance of modern computers could make iterative image reconstruction fast enough to be viable, provided we are able to deal with the large number of probability coefficients for the system response matrix in high-resolution PET scanners, which is a difficult task that prevents the algorithms from reaching peak computing performance. Considering all possible axial and in-plane symmetries, as well as certain quasi-symmetries, we have been able to reduce the memory requirements to store the system response matrix (SRM) well below 1 GB, which allows us to keep the whole response matrix of the system inside RAM of ordinary industry-standard computers, so that the reconstruction algorithm can achieve near peak performance. The elements of the SRM are stored as cubic spline profiles and matched to voxel size during reconstruction. In this way, the advantages of on-the-fly calculation and of fully stored SRM are combined. The on-the-fly part of the calculation (matching the profile functions to voxel size) of the SRM accounts for 10 to 30% of the reconstruction time, depending on the number of voxels chosen. We tested our approach with real data from a commercial small animal PET scanner. The results (image quality and reconstruction time) show that the proposed technique is a feasible solution.
{
"annotation_id": "fcc1f105-5ed6-405e-a160-eed8b289faae",
"date_created": "2026-03-02T18:01:11.175000Z",
"date_modified": "2026-03-02T18:01:11.175000Z",
"file_hash": "99f9cb9c7e481f6be4bd83c1ac2080007ffe089d197a366a70fbf15c06bdf4ee",
"private": false,
"record": {
"abstract": "Small animal PET scanners require high spatial resolution and good\nsensitivity. To reconstruct high-resolution images in 3D-PET, iterative\nmethods, such as OSEM, are superior to analytical reconstruction algorithms,\nalthough their high computational cost is still a serious drawback. The higher\nperformance of modern computers could make iterative image reconstruction fast\nenough to be viable, provided we are able to deal with the large number of\nprobability coefficients for the system response matrix in high-resolution PET\nscanners, which is a difficult task that prevents the algorithms from reaching\npeak computing performance. Considering all possible axial and in-plane\nsymmetries, as well as certain quasi-symmetries, we have been able to reduce\nthe memory requirements to store the system response matrix (SRM) well below 1\nGB, which allows us to keep the whole response matrix of the system inside RAM\nof ordinary industry-standard computers, so that the reconstruction algorithm\ncan achieve near peak performance. The elements of the SRM are stored as cubic\nspline profiles and matched to voxel size during reconstruction. In this way,\nthe advantages of on-the-fly calculation and of fully stored SRM are combined.\nThe on-the-fly part of the calculation (matching the profile functions to voxel\nsize) of the SRM accounts for 10 to 30% of the reconstruction time, depending\non the number of voxels chosen. We tested our approach with real data from a\ncommercial small animal PET scanner. The results (image quality and\nreconstruction time) show that the proposed technique is a feasible solution.",
"arxiv_id": "physics/0609104",
"authors": [
"J. L. Herraiz",
"S. Espana",
"J. J. Vaquero",
"M. Desco",
"J. M. Udias"
],
"categories": [
"physics.med-ph",
"physics.ins-det"
],
"doi": "10.1088/0031-9155/51/18/007",
"journal_ref": "Physics in Medicine and Biology, Volume 51, Number 18, 21\n September 2006",
"title": "FIRST: Fast Iterative Reconstruction Software for (PET) tomography",
"url": "https://arxiv.org/abs/physics/0609104"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "48757ecd-a4ed-4f4f-af88-575a718a271b",
"id": "arXiv Dataset IDs",
"type": "Model",
"variant": "snapshot-2026-03-01",
"version": "0.1.0"
},
"user_id": 1000002
}