Machine learning volumetry of ischemic brain lesions on CT after thrombectomy-prospective diagnostic accuracy study in ischemic stroke patients.

Purpose: Ischemic lesion volume (ILV) is an important radiological predictor of functional outcome in patients with anterior circulation stroke. Our aim was to assess the agreement between automated ILV measurements on NCCT using the Brainomix software and manual ILV measurements on diffusion-weighted imaging (DWI).

Methods: This was a prospective single-center observational study of patients with CT angiography (CTA) proven anterior circulation occlusion treated with endovascular thrombectomy (May 2018 to May 2019). NCCT ILV was measured automatically by the Brainomix software. DWI ILV was measured manually. The McNemar's test was used to test sensitivity and specificity. The Somer's delta was used to test the differences between concordant and discordant ASPECTS regions. The Bland-Altman plot was calculated to compare the differences between Brainomix and DWI ILVs.

Results: Forty-five patients were included. Median Brainomix ILV was 23 ml (interquartile range [IQR], 15-39 ml), and median DWI ILV was 11.5 ml (IQR, 7-32 ml) in the TICI 2b-3 group. In the TICI 0-2a, the NCCT ILV was 39 ml (IQR, 18-62 ml) and DWI ILV was 30 (IQR, 11-105 ml). The DWI ILVs in patients with good clinical outcome (mRS 0-2) was significantly lower compared with patients with mRS ≥ 3 (10 mL vs 59 mL, p = 0.002). Similar trend was observed for Brainomix ILV measurements (21 mL vs 39 mL, p = 0.012). There was a high correlation and accuracy in the detection of follow-up ischemic changes in particular ASPECTS regions.

Conclusion: NCCT ILV measured automatically by the Brainomix software might be considered a valuable radiological outcome measure.