6533b86efe1ef96bd12cc047

RESEARCH PRODUCT

Left-Ventricle Segmentation of SPECT Images of Rats

Alain LalandeAlexandra OudotPierre-marc JodoinFredy Pinheiro

subject

[SDV.IB.IMA]Life Sciences [q-bio]/Bioengineering/ImagingHeart VentriclesBiomedical Engineering030204 cardiovascular system & hematologySingle-photon emission computed tomography030218 nuclear medicine & medical imaging03 medical and health sciences0302 clinical medicineImaging Three-DimensionalCutmedicineAnimalsComputer visionSegmentationLongitudinal Studies[ SDV.IB.IMA ] Life Sciences [q-bio]/Bioengineering/ImagingComputingMilieux_MISCELLANEOUSMathematicsTomography Emission-Computed Single-Photonmedicine.diagnostic_testCardiac cycleOrientation (computer vision)business.industryImage segmentationRatsTomographyArtificial intelligencebusinessEmission computed tomography

description

Single-photon emission computed tomography (SPECT) imaging of the heart is helpful to quantify the left-ventricular ejection fraction and study myocardial perfusion scans. However, these evaluations require a 3-D segmentation of the left-ventricular wall on each phase of the cardiac cycle. This paper presents a fast and interactive graph cut method for 3-D segmentation of the left ventricle (LV) of rats in SPECT images. The method is carried out in three steps. First, 3-D sampling of the LV cavity is made in a spherical-cylindrical coordinate system. Then, a graph-cut-based energy minimization procedure provides delineation of the myocardium centerline surface. From there, it is possible to outline the epicardial and endocardial boundaries by considering the second statistical moment of the SPECT images. An important aspect of our method is to always produce anatomically coherent U-shape results. It also relies on only two intuitive parameters regulating the smoothness and the thickness of the segmentation result. Results show not only that our method is statistically as accurate as human experts, but it is one order of magnitude faster than a state-of-the-art method with a processing time of at most 2 s on a 4-D cardiac image after having determined the LV orientation.

10.1109/tbme.2015.2422263https://hal.archives-ouvertes.fr/hal-01174039