Computer-assisted system for diagnosing degenerative dementia using cerebral blood flow SPECT and 3D-SSP: a multicenter study

Abstract

Purpose: Due to increasing numbers of patients with dementia, more physicians who do not specialize in brain nuclear medicine are being asked to interpret SPECT images of cerebral blood flow. We conducted a multicenter study to determine whether a computer-assisted diagnostic system Z-score summation analysis method (ZSAM) using three-dimensional stereotactic surface projections (3D-SSP) can differentiate Alzheimer's disease (AD)/dementia with Lewy bodies (DLB) and non-AD/DLB in institutions using various types of gamma cameras.

Method: We determined the normal thresholds of Z-sum (summed Z-score) within a template region of interest for each single photon emission computed tomography (SPECT) device and then compared them with the Z-sums of patients and calculated the accuracy of the differential diagnosis by ZSAM. We compared the diagnostic accuracy between ZSAM and visual assessment.

Patients: We enrolled 202 patients with AD (mean age, 76.8 years), 40 with DLB (mean age 76.3 years) and 36 with non-AD/DLB (progressive supranuclear palsy, n = 10; frontotemporal dementia, n = 20; slowly progressive aphasia, n = 2 and one each with idiopathic normal pressure hydrocephalus, corticobasal degeneration, multiple system atrophy and Parkinson's disease) who underwent N-isopropyl-p-[(123)I] iodoamphetamine cerebral blood flow SPECT imaging at each participating institution.

Results: The ZSAM sensitivity to differentiate between AD/DLB and non-AD/DLB in all patients, as well as those with mini-mental state examination scores of ≥24 and 20-23 points were 88.0, 78.0 and 88.4 %, respectively, with specificity of 50.0, 44.4 and 60.0 %, respectively. The diagnostic accuracy rates were 83.1, 72.9 and 84.2 %, respectively. The areas under receiver operating characteristics curves for visual inspection by four expert raters were 0.74-0.84, 0.66-0.85 and 0.81-0.93, respectively, in the same patient groups. The diagnostic accuracy rates were 70.9-89.2 %, 50.9-84.8 % and 76.2-93.1 %, respectively.

Conclusion: The diagnostic accuracy of ZSAM to differentiate AD/DLB from other types of dementia or degenerative diseases regardless of severity was equal to that of visual assessment by expert raters even across several institutions. These findings suggested that ZSAM could serve as a supplementary tool to help expert evaluators who differentially diagnose dementia from SPECT images by visual assessment.