Validation of the cingulate island sign with optimized ratios for discriminating dementia with Lewy bodies from Alzheimer’s disease using brain perfusion SPECT

Dementia with Lewy bodies (DLB) is often cited as the second most common dementia after Alzheimer’s disease (AD). It is clinically important to distinguish DLB from AD because specific side effects of antipsychotic drugs are limited to DLB. The relative preservation of glucose metabolism in the posterior cingulate gyri versus that in the precuni, the cingulate island sign (CIS) [1], was reported in 1997 by Imamura et al. [2] to be higher in patients with DLB than those with AD, and is highly specific for diagnosing DLB. Lim et al. [1] reported that the sensitivities for using CIS to diagnose DLB ranged from 62 to 86% and only 43 to 50% for hypometabolism in the medial occipital lobe. Recently, Graff-Radford et al. [3] reported the pathologic association of CIS with autopsy findings. These early CIS studies were limited to ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG-PET) imaging. O’Brien et al. [4] showed that, even with FDG, the differentiation accuracy was 72.4%, and furthermore that the CIS was not observed in SPECT images in their study. Using brain perfusion SPECT to differentiate DLB from AD [5], we previously showed that the largest value for the area under the receiver operating characteristic (ROC) curve (AUC) was 0.873; the accuracy, sensitivity, and specificity were 85.7, 88.9, and 82.4%, respectively, obtained for the ratio of the posterior cingulate gyri within early Alzheimer’s disease-specific hypoperfusion volumes of interest (VOI) in a group comparison between patients with AD and cognitively normal subjects versus the medial occipital lobe. The AUC calculated using the medial occipital VOIs was only 0.614, while the accuracy, sensitivity, and specificity were 68.6, 55.6, and 82.4%, respectively. Because AD and DLB are known to have pathological overlaps and it is difficult to distinguish these two diseases using ¹⁸F-FDG-PET or brain perfusion SPECT [4], it was logical to use these early AD-specific VOIs to discriminate AD from DLB [5].

Therefore, in the current study to optimize these VOIs and to verify the clinical usefulness of CIS in different groups of patients [i.e., group A with DLB from our previous study, group B with DLB from the present study and group C with AD from the present study (Table 1)], we first compared a group of patients with DLB (group A in Table 1) to normal controls to obtain specific VOIs. We then applied these optimized VOIs to new groups (group B and C in Table 1) of patients with DLB and AD to evaluate their accuracy using ROC curve analysis and to determine the thresholding CIS values to discriminate DLB from AD.

Accordingly, two purposes of this study are as follows: one is optimization of VOI setting for calculating CIS values by setting statistically significant hypoperfusion areas as VOIs for both denominators and numerators and the other is to evaluate their accuracy and simultaneously to retest the method described in our previous paper [5].

The patient characteristics are summarized in Table 1. No significant differences were observed for age or gender. All the heart to mediastinum ratios in the [¹²³I] MIBGs scanned 3 h after injection in patients with DLB were below 2.0.

As shown in Table 2, there were no significant differences among any of the analysed AUCs. Combinations of these that resulted in AUCs ≥0.87 are listed in Table 3. A higher accuracy of 84.6% was obtained with three VOI combinations for numerator and denominator when the ratio sum of all the positive Z scores within those VOIs was used as thresholds. The first of these utilized the optimized VOIs: tAD_DLB_num_2/tDLB_specific_VOI_1, which was termed as “CISRo” (CIS ratio optimized); the second the mixed VOIs: tAD_DLB_num_1/tAAL_PreC&C, which was termed as “CISRmix” (CIS ratio optimized and with AAL); and the third the AAL VOIs: tAAL_PCG/tAAL_medOccipital, which was termed as “CISRaal” (CIS ratio with AAL). The accuracy, sensitivity, and specificity of the CISRo for differentiating patients with DLB from those with AD were 84.6, 92.3, and 76.9%, respectively, with a threshold value less than 0.281; for the CISmix, they were 84.6, 84.6, and 84.6%, respectively, with a threshold value less than 0.348; and for the CISRaal, they were 84.6, 76.9, and 92.3%, respectively, with a threshold value less than 0.0308. Box plots of these CIS ratios with optimized VOIs and AAL VOIs were in Fig. 2.

The CIS, the relative preservation demonstrated in ¹⁸F-FDG-PET of glucose metabolism in the posterior cingulate gyri versus that in the precuni, in patients with DLB compared with those with AD is considered to be highly specific for diagnosing DLB [1]. In our previous study [5] of brain perfusion SPECT using [^99mTc] ECD, we also observed a differential CIS ratio when the method was optimized using the posterior cingulate area of an early AD-specific VOI included in eZIS software as numerator and the anatomical medial occipital area as denominator for this ratio. Because brain perfusion SPECT is relatively more economical and widely available worldwide than PET, the clinical application of CIS to differentiate patients with DLB from AD would be more effective using brain perfusion SPECT. In the current study, not only did we determine the threshold value for clinical use, but we also tried to optimize the CIS in brain perfusion SPECT and at the same time to retest it in another patient cohort. As a result, we obtained a larger AUC than that of our previous study [5].

Previously, we used the posterior cingulate area within the early AD-specific VOI as numerator, and did not take into account a DLB-specific VOI. When we used the same numerator in the current retest study that previously produced the largest AUC, we received an AUC of 0.858. Although this was smaller than the previous AUC value of 0.873, it still fell within the 95% confidence interval. Also the accuracy (84.6 versus 85.7%) and sensitivity (84.6 versus 88.9%) of the retest were slightly lower than those of the previous study; however, also all of these values are over 84% and thus appear to be reliable enough to evaluate the differentiation of DLB from AD.

In the original CIS study [1], AAL_PCG VOI was used as a numerator and AAL_PreC&C VOI as a denominator, resulting in accuracy, sensitivity, and specificity of 78, 77, and 80%, respectively. These ratios were also used as thresholds, and ROC analysis was applied in the current study, which resulted in accuracy, sensitivity, and specificity of 76.9, 61.5, and 92.3%, respectively. These results indicate that even though the resolution of SPECT is generally inferior to those of PET, the accuracy, sensitivity, and specificity obtained in the current study were comparable to those obtained with PET. Furthermore, we also effectively optimized the VOI parameters.

DLB is often cited as the second most common dementia after AD. It is clinically important to distinguish DLB from AD because specific side effects of antipsychotic drugs are limited to DLB. In discriminating DLB from AD clinically, dopamine transporter (DAT) imaging [17] and [¹²³I] MIBG [18] are useful because they detect early disturbances of the nigrostriatal pathway or peripheral sympathetic nervous system in patients with DLB. With a combination of these two techniques, over 90% sensitivity and specificity in discriminating DLB from AD are reported [19]. Nevertheless, brain perfusion SPECT is more widely and commonly used for clinical screening of patients with dementia. Furthermore, compared with morphometric imaging, SPECT is a more sensitive modality for functional imaging used to detect early stages of neurodegenerative disease before shrinkage [20]. SPECT also reveals useful information for differentiating AD as well as other dementias, including vascular dementia or frontotemporal lobe degeneration [21].

In this study, in trying to optimize the use of the CIS in SPECT, we subtracted DLB-specific VOIs derived from comparisons between DLB and normal controls from AD-specific VOIs. Furthermore, the CIS ratio was determined by comparing these with specific hypoperfusion areas in DLB instead of the entire anatomical occipital area. This resulted in a higher AUC of 0.882 compared with our previous study [21]. With these optimized numerators and denominators determining the CIS ratio, we obtained the largest specificity of 92.3% for differentiating DLB from AD. Hypothetically, if we used this procedure with highly specific examinations such as [¹²³I] MIBG [22] or DAT imaging [19], an even higher accuracy might be achieved. Moreover, a combination of these two with brain perfusion SPECT might prove to be even more valuable, considering the purpose of the screening, compared with using only MIBG and DAT, which only detect degeneration of central or peripheral monoamine pathways, while brain perfusion SPECT measures whole brain activity and can differentiate many brain disorders.

Contrary to practical clinical situations, amyloid PET was used as a biomarker for diagnosing AD in this study. As a subject of important future investigation, additional clinical prospective study using our CIS procedure for diagnosing AD without amyloid PET may reinforce the effective diagnostic flowchart with combination of the CIS, amyloid PET, MIBG and DAT imaging.

The CIS observed in ¹⁸F-FDG-PET was reportedly indicative of a lower Braak neurofibrillary tangle stage in patients with DLB [3]. In our study, we compared the CIS detected using brain perfusion SPECT in patients with DLB and AD. While we cannot compare these two studies directly, brain perfusion and metabolism are physiologically coupled [23]. The limitations of SPECT include lower image resolution and a large partial volume effect. These limitations, in conjunction with a patient’s pathological status, result in decreased accuracy when attempting to differentiate between DLB and AD. However, optimization of the procedure using the CIS ratio described in this paper seems to overcome these limitations and achieve a larger accuracy compared with the existing procedure. Furthermore, SPECT is widely accessible and more economical than PET.

To further the clinical usage of CIS to discriminate between DLB from AD, we retested and optimized the procedure, and thereby achieved a larger AUC of 0.882. In so doing, the accuracy, sensitivity, and specificity were 84.6, 92.3, and 76.9%, respectively, with a threshold value less than 0.281 for CISRo, 84.6, 84.6, and 84.6%, respectively, with a threshold value less than 0.348 for CISRmix, and 84.6, 76.9, and 92.3%, respectively, with a threshold value less than 0.03 for CISRaal. Brain perfusion SPECT is not only useful in screening for dementia and economically and widely accessible, but also can facilitate the differentiation of DLB from AD when these new CIS ratios are properly used.