Optimization of the Attenuation Coefficient for Chang Attenuation Correction in ¹²³I Brain Perfusion SPECT

DISCUSSION

We determined μ_opt-values in ChangAC and ChangACSC retrospectively using clinical data. The μ_opt-values were 0.140 for ChangAC and 0.160 for ChangACSC. However, some brain regions were under- or overestimated in the SPECT images when the μ_opt-values were applied. Limitations of the Chang method, as a uniform AC, were also revealed.

The distribution of μ_opt-values per patient displayed some variation, with 0.140 being the most common value for ChangAC and 0.160 and 0.165 being most common for ChangACSC. However, it is not practical to apply individual μ-values for each patient. Therefore, we determined μ_opt-values by averaging out the variation in patients by analyzing all brain regions together in 60 reference patients. By comparing the counts of the Chang method with those of the CT-based method, we identified significant differences for all μ-values except 0.140 in ChangAC and 0.160 and 0.165 in ChangACSC, thereby providing statistical support for the μ_opt-value. Stodilka et al. (12) reported a relative quantification error of 20% by applying a μ-value of 0.120 to ChangAC in a phantom study. In the present patient-based study, the absolute percentage error for ChangAC applying the μ_opt-value was 22.81%. Although the degree of error was comparable, the μ_opt-values for phantoms and patients were different. For quantitation using ¹²³I, Iida et al. reported a maximum percentage error of 30% for ChangAC with a μ-value of 0.090 and for ChangACSC with a μ-value of 0.166 (9). For ChangAC, a smaller percentage error was achieved by applying the μ_opt-value in this study. For ChangACSC, there was no noticeable difference from this study because the μ-values were close to our μ_opt-values. Interindividual variability of anatomic standardization in brain perfusion SPECT is 3%–9% (21–23). The absolute percentage error between the CT-based and Chang methods in the present study averaged 4.45% for ChangAC and 6.37% for ChangACSC. Application of the μ_opt-values achieved an AC error comparable to the interindividual variation of anatomic standardization that can occur in routine clinical practice.

Bland–Altman analysis revealed systematic bias in ChangAC and ChangACSC when applying the μ_opt-value, and agreement with the CT-based method was not perfect. The percentage error for ChangAC and ChangACSC tended to be underestimated in the limbic brain regions and overestimated in the central brain regions. This trend was visually confirmed in an example of the SPECT image of a reference patient shown in Figure 4. Ito et al. (4) reported an overestimation of central brain regions in ChangACSC applying a μ-value of 0.166, relatively close to our μ_opt-value. These facts highlight the limitations of the Chang method even when using μ_opt-values and reiterate the superiority of the CT-based method.

The skull possesses a higher μ-value than brain tissue because of greater photon loss. Further, the thickness of the skull varies slightly with age and sex (24). The different μ_opt-values per patient in the present study appear to be due to skull thickness variations among patients. Nicholson et al. (11) reported that the skull paradoxically affects broad-beam μ-values, and other studies (12,13) also reported lower μ-values than those of uniform soft tissue. Stodilka et al. (12) observed that μ_opt-values at the cerebellar level, surrounded by thick bony structures, are smaller than those at the basal ganglia level. Iida et al. (9) reported smaller μ-values at the cerebellar level, where the airway is included in the slice position, and higher μ-values at the parietal level, where the skull is relatively thick. These reports indicate a complex interplay of several factors involved in optimizing μ-values. In this study, we averaged the differing μ_opt-values in patients by a combined analysis of 60 reference patients and accounted for differences in correction error among brain regions by analyzing all brain regions together using anatomic standardization. Our method provides a generalizable μ_opt-value that considers differences in skull thickness (an error factor for interpatient variation) and differences in attenuation structure for each imaging slice position (an error factor for intrapatient variation). Consequently, the μ_opt-value determined in the present study is relatively high compared with the various μ-values conventionally used.

The theoretic narrow-beam μ-value for water for the 159-keV γ rays emitted by ¹²³I is 0.148 (25). Using the theoretic narrow-beam μ-value for ChangAC overcorrects for attenuation and overestimates the brain center region (26). Harris et al. (27) reported that a slightly lower μ-value should be applied than the theoretic μ-value. Here, the μ_opt-value of ChangAC was also lower than the theoretic narrow-beam μ-value.

The μ_opt-value of ChangACSC in our study was higher than the theoretic narrow-beam μ-value, whereas previous studies proposed lower values than the theoretic narrow-beam μ-value (12,13,16). Some previous studies using phantoms focused on assessing the uniformity of AC (12,16) because brain perfusion SPECT images are commonly normalized by the maximum count in the series. The uniformity of AC contributes to qualitative improvement. In this study using clinical data, the SD of the percentage error at low μ-values was small, and uniformity within the series was preserved. Contrarily, the absolute percentage error applying μ_opt-values was comparable to that reported by Stodilka et al. (12), indicating that quantification was assured. Iida et al. (9) reported no apparent difference in regional CBF images obtained with the measured attenuation map and ChangACSC applying a high μ-value of 0.166. Therefore, it is difficult for the Chang method to achieve both qualitative and quantitative performance because low μ-values contribute to qualitative improvement whereas high μ-values contribute to quantitative improvement. Since the degree of contribution of high μ-values to the quantitation improvement was greater than the degree of contribution of low μ-values to the qualitative improvement, using high μ-values is recommended.

We validated the μ_opt-value determined using the reference group in hypoperfusion regions to confirm their adaptability to the disease group. The choice of target diseases was considered so that the entire brain region (anterior, posterior, lateral, parietal, and basal regions) could be included as hypoperfusion regions. We observed no significant difference in the percentage error in the hypoperfusion regions in the reference and disease groups when the μ_opt-value was applied, indicating adaptability of the μ_opt-value for the disease group.

The reference group used in the present study included patients who underwent routine clinical examinations and not healthy volunteers. However, conducting studies on healthy volunteers is not practical to determine μ_opt-values. Licho et al. (28) also evaluated patients who underwent routine clinical examinations to validate the effects of various AC methods. In clinical routine, there are few patients in whom brain perfusion is generally preserved. We included patients diagnosed with a mild degree of nonspecific hypoperfusion in the reference group to obtain a larger cohort of patients for inclusion in the present study. Additionally, including more patients enables more generalizable μ_opt-values to be determined.

This study had some limitations. First, it was difficult for all patients to achieve an ideal head tilt because of patient-specific limitations in body position during SPECT imaging. These factors may have influenced the variation of μ_opt-values depending on the slice position. Second, the specifications for SC of the SPECT/CT system used in this study were limited to the dual-energy window method only. Since ¹²³I also emits photons with energies of as high as 529 keV, it is best to use a multiple-window method, including the triple-energy-window method, to improve the effects of down scatter. Third, we investigated using one type of γ-camera for ¹²³I with constant parameters for image reconstruction and SC. A preliminary validation using several patients confirmed that the μ_opt-values for each patient did not change when the number of iterations for image reconstruction and the weighting factor for SC was changed; however, this was not sufficient. The possibility that μ_opt-values may change when other radionuclides, γ-cameras, or other SC are used cannot be extrapolated to other clinical applications. These should be investigated in further studies.