In this proof-of-concept study, we have been able to show that it is not only feasible to use automated tools to have strong internal consistency with hematoma volume calculation, but also reliably assess for HE on serial imaging. While identification of ICH has been shown to be achievable with automated software,11–13 this is the first study to add a new dimension and FDA-cleared volume calculation that can be used for assessment of HE across serial imaging studies. Specifically, in regards to HE, we show a fairly high sensitivity and specificity in this pilot data using already established definitions of HE in the literature.
Though focused on addressing the feasibility of this approach, this analysis was strengthened by using real-world cases as well as having all manual measurements conducted by one reviewer. This methodology helped strengthen the generalizability of our results as well as eliminated any potential inter-rater reliability within hematoma measurements. Similarly, the inclusion of RCT scans was conducted to help attain an appropriate sample size for this analysis. There is a potential confounder with the impact of the therapeutic intervention on HE, however our analysis was not interested in the absolute rates of HE but rather the ability to identify HE on serial imaging.
Another aspect of our approach was the identification of patients that exhibited a ‘negative expansion,’ or patients where the difference between baseline hematoma volume and the subsequent volume resulted in a negative value. These cases routinely occurred with ICHs in the brainstem where slice thickness and image acquisition can result in slight differences in which slices show most of the hematoma. However, manual review of these cases failed to result in the identification of a patient with true HE that was masked due to issues of imaging acquisition.
The generalizability of these results stems from the definitions of HE used for this analysis. There is no consistency in the literature regarding how HE is defined,14 with some studies using an absolute volume increase threshold (i.e. 6cc increase) or percentage increase threshold (i.e. 33% increase) or a combination of both rules. Additionally, there is no consensus on a threshold that qualifies as a clinically meaningful increase, only that the presence of HE suggests a poorer prognosis.5–7 Given the inconsistency in the field, we specifically choose the most widely used thresholds and used both rules in conjunction to help increase the specificity of identification.
When applying our method of HE to clinical management, our approach decreases false positive identification and would potentially highlight the highest risk patients showing expansion. With new automated tools being developed to predict HE,16 we will need to have concurrent tools to reliably identify HE in the field. Automated software can potentially serve to appropriately highlight patients with expansion in order to tailor therapeutic interventions or triage patients for closer monitoring. Especially in busy medical centers, automated software can aid in the rapid identification of these at-risk patients as well as provide prognostic insight that might not be achievable without specialized subspecialty care.