This study demonstrated that the reproducibility of PCT measurements, including CBV, CBF, MTT, and BBBP values, as well as of PCT abnormality volumes based on CBV and MTT thresholds, was improved when an automated post-processing algorithm minimizing user interaction was used. The average interobserver variability for CBV, CBF and MTT measurements using an automated post-processing approach was 10.4%, almost three times less than the interobserver variability of 27.1% observed for a manual post-processing approach. This improved interobserver reproducibility when using an automated approach was observed both in the ischemic parenchyma (10.1% with automated post-processing, 24.4% with manual post-processing) and in the non-ischemic parenchyma (10.8% with automated post-processing, 29.8% with manual post-processing). In agreement with previous studies [15
], our results showed that MTT was the most reproducible PCT parameter.
In this study, we reported not just the variability of measurements of CBV, CBF, MTT, and BBBP, but also the variability of measurements of infarct core and total perfusion defect volumes based on CBV and MTT thresholds. To our knowledge, the interobserver variability for measurements of infarct core and total perfusion defect volumes has not been reported before. The interobserver variability we obtained for the measurements of CBV, CBF and MTT using a manual processing approach is similar to what was reported in a prior published study comparing manual post-processing of PCT datasets by different CT technologists [16
]. In another study, lower interobserver variability in case of manual PCT post-processing was achieved by the use of a uniform and standardized approach [17
]. In the same study, paradoxically, greater interobserver variability was introduced with the use of an automated approach [17
]. This reflects that not all automated PCT post-processing algorithms, as implemented in commercially available packages, are equivalent, and that each of them needs to be carefully evaluated, and possibly improved.
Since a completely automated approach to post-processing PCT data would yield a perfect degree of agreement (0% of interobserver variability), it is clear that the interobserver variability around 10.4% observed when using the automated approach was introduced by the user-defined manual adjustments. The manual adjustments performed by the observers in our study were mainly justified by an incorrect selection of the AIF and VOF by the automatic algorithm in studies with severe head motion artifact. Although these manual corrections were necessary to obtain accurate PCT results, they introduced interobserver variability for the “automated” PCT results. This again emphasizes the need for further improvement of automated PCT post-processing algorithms, with the ultimate goal being to completely eliminate user interaction.
Two main limitations should be noted in our study. The first one is that the results we obtained apply only to a specific software implementation. Therefore, the variability between automated and manual post-processing observed in our study should not be assumed to be similar to other automated PCT post-processing algorithms. The second limitation is the absence of a well-defined criterion of acceptability in the variability of PCT-derived measurements. However, prior studies with similar results [16
] have shown that an overall variability around 10% does not change the qualitative visual assessment of PCT maps, thus being very unlikely to alter clinical management decisions.
In conclusion, automated post-processing of PCT data improves interobserver variability in measurements of CBV, CBF, MTT and BBBP, as well as volume of infarct core and penumbra based on CBV and MTT thresholds. PCT post-processing automation is a way to ensure reliability of PCT technique when widely used in the community, which is one of the requirements for it to be used in the selection of stroke patients for acute reperfusion therapies.