Premium
P2–379: Memory function tests correlate differently to regional cerebral blood flow in mild cognitive impairment and Alzheimer's disease. A 123 I–IMP SPECT study with statistical parametric mapping
Author(s) -
Ansar Md. Ashik B.,
Kazui Hiroaki,
Oku Naohiko,
Kimura Yasuyuki,
Kimura Nobuyo,
Kato Hiroaki,
Nessa Begum N.,
Ikejiri Yashitaka,
Takeda Masatoshi,
Hatazawa Jun
Publication year - 2006
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2006.05.1219
Subject(s) - cerebral blood flow , nuclear medicine , wechsler adult intelligence scale , statistical parametric mapping , single photon emission computed tomography , emission computed tomography , psychology , correlation , alzheimer's disease , medicine , cognition , cardiology , positron emission tomography , neuroscience , radiology , magnetic resonance imaging , disease , mathematics , geometry
achieve a uniform quality level for this study, the dataset was reduced to 224 subjects from 6 centers. A supervised fully connected ANN was implemented with a single layer of hidden units and 30 inputs (3 modalities, each with their respective 3x3 neighborhood and spatial information). The output comprised of four probability maps GM, WM, CSF and ARWMC, but only the ARWMC output is assessed here. Before segmentation all data were pre-processed for head motion and intensity variations. The ANN was trained with an ROI dataset drawn on 6 subjects from 4 centers. For segmentation, the optimal threshold of the ARWMC probability map was determined using the Index of Similarity describing the overlap between ANN and manual ROI’s. Here a threshold of 90% was found to be optimal. Results: From each of the 6 center the correlation coefficient between ANN and manually determined were generally high: 0.97, 0.97, 0.87, 0.88, 0.73 and 0.98. The correlation of the combined set was 0.89. The two first centers, with a correlation coefficient of 0.97, had two subjects in the training set which may explain their better performance. Deviating quality in the FLAIR image contributes to the lowest correlation in a center. In other cases a systematic bias was present in some anatomical regions (e.g. septum pellucidum, where lesion were not drawn manually) or regarding lacunar infarcts. Conclusion: In this multi-center investigation of ARWMC segmentation a high average correlation of 0.89 was found. Improvements may be gained from representing more centres in the training set. The results demonstrate the generalizability of ANN methods, and underline the importance of rigorous standardization of MRI quality in multi-center studies.