Affiliations: [a]
Istituto Nazionale di Fisica Nucleare, Sezione di Genova, Genova, Italy
| [b] Dipartimento di Fisica, Università degli Studi di Genova, Genova, Italy
| [c] Department of Nuclear Medicine, Fondazione Poliambulanza, Brescia, Italy
| [d]
IRCCS Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy
| [e] Department of Medical and Experimental Sciences, Unit of Neurology, Brescia University, Brescia, Italy
| [f]
University Hospitals and University of Geneva, Geneva, Switzerland
| [g] Nuclear Medicine Unit, Department of Health Sciences (DISSAL), Genoa University and IRCCS AOU S. Martino-IST, Genova, Genova, Italy
| [h] Clinical Neurobiology Unit, Department of Neuroscience (DINOGMI), Genoa University and IRCCS AOU S. Martino-IST, Genova, Genova, Italy
Correspondence:
[*]
Correspondence to: Andrea Chincarini, INFN, via Dodecaneso 33, I-16146 Genova, Italy. Tel.: +39 010 353 6496; Fax: +39 010 313358; E-mail: [email protected].
Abstract: The assessment of in vivo 18F images targeting amyloid deposition is currently carried on by visual rating with an optional quantification based on standardized uptake value ratio (SUVr) measurements. We target the difficulties of image reading and possible shortcomings of the SUVr methods by validating a new semi-quantitative approach named ELBA. ELBA involves a minimal image preprocessing and does not rely on small, specific regions of interest (ROIs). It evaluates the whole brain and delivers a geometrical/intensity score to be used for ranking and dichotomic assessment. The method was applied to adniimages 18F-florbetapir images from the ADNI database. Five expert readers provided visual assessment in blind and open sessions. The longitudinal trend and the comparison to SUVr measurements were also evaluated. ELBA performed with area under the roc curve (AUC) = 0.997 versus the visual assessment. The score was significantly correlated to the SUVr values (r = 0.86, p < 10-4). The longitudinal analysis estimated a test/retest error of ≃2.3%. Cohort and longitudinal analysis suggests that the ELBA method accurately ranks the brain amyloid burden. The expert readers confirmed its relevance in aiding the visual assessment in a significant number (85) of difficult cases. Despite the good performance, poor and uneven image quality constitutes the major limitation.
Keywords: Alzheimer’s disease, amyloid, image analysis, mild cognitive impairment, PET, standardized uptake value ratio
