An automated ROI setting method using NEUROSTAT on cerebral blood flow SPECT images

Ann Nucl Med. 2009 Jan;23(1):33-41. doi: 10.1007/s12149-008-0203-7. Epub 2009 Feb 11.


Objective: We have developed a method to automatically set regions of interest (ROI) (automated ROI) on cerebral blood flow single-photon emission computed tomography (SPECT) images with morphological information specific to the subjects. The objective was to set ROIs automatically without losing individual morphological information in the SPECT images and then evaluate its validity and clinical applicability.

Methods: We constructed the volume of interest (VOI) template on the standardized brain generated by NEUROSTAT to determine the regions for ROIs to be set. Assuming patients with cerebral vascular disease, the VOI template was constructed so that the ROIs were drawn for the major vascular regions and 17 regions in total within the hemisphere, basal ganglia, thalamus, cerebellar cortex, cerebellar vermis, and pons. By comparing the major vascular occlusion models, the accuracy of region setting by the VOI template was evaluated for validation. Using the anatomical standardization of NEUROSTAT and inverse transformation, the automated ROI transformed the VOI template into the individual brain shape and then the VOI template was extracted from each slice to determine ROIs. An evaluation was made by visually investigating the effect of a different image quality and cerebral blood flow tracers using brain phantom and clinical data. The regional cerebral blood flow (rCBF), determined by the manual setting method of ROI (manual ROI) and automated ROI, was compared. We also compared automated ROI with other morphological images using clinical data.

Results: The VOI templates accurately showed the region with the reduced blood flow in the major vascular occlusion model, which validated the proper ROI setting. The brain phantom study demonstrated that ROI settings were least influenced by matrix size, image quality, and image rotation. The observation with the clinical data also indicated that the variation in cerebral blood flow tracers little affected the ROI settings. The comparison with manual ROI revealed a strong correlation between the two ROI settings, and the mean values within both ROIs were similar. The comparative evaluation with morphological images, obtained by magnetic resonance imaging (MRI), verified the accurate setting of ROI.

Conclusions: The automated ROI achieved successful automatic ROI settings without distorting individual SPECT images. The automated ROI is not affected by the differences in the image quality or the cerebral blood flow tracers, which suggests versatile applicability. Thus, the use of automated ROI may eliminate the interoperator and interfacility variability in ROI setting and improve objectivity and reproducibility. It also allows comparative evaluation at the same transverse level with images acquired with other modalities such as MRI and is expected to enhance the clinical diagnosis.

Publication types

  • Evaluation Study

MeSH terms

  • Algorithms
  • Artificial Intelligence
  • Blood Flow Velocity / physiology*
  • Brain / blood supply
  • Brain / diagnostic imaging*
  • Brain / physiology*
  • Cerebrovascular Circulation / physiology*
  • Humans
  • Image Enhancement / methods
  • Image Interpretation, Computer-Assisted / methods*
  • Pattern Recognition, Automated / methods*
  • Phantoms, Imaging
  • ROC Curve
  • Reproducibility of Results
  • Sensitivity and Specificity
  • Tomography, Emission-Computed, Single-Photon / methods*