Deep learning-based computed tomography reconstruction improves image quality but does not significantly affect Alberta stroke program early CT score evaluation in acute middle cerebral artery territory infarction.

PURPOSE: To evaluate whether deep learning image reconstruction (DLIR), a deep learning-based computed tomography (CT) reconstruction method, improves image quality, Alberta Stroke Program Early CT Score (ASPECTS), and the detectability of early ischemic changes (EIC) in acute middle cerebral artery (MCA) territory infarction compared to filtered back projection (FBP) and Iterative Reconstruction (IR).

METHODS

This retrospective study included 30 patients with confirmed acute MCA infarction who underwent non-contrast brain CT and follow-up diffusion-weighted imaging (DWI). CT images reconstructed with FBP, IR, and DLIR were assessed for image quality using quantitative metrics (signal-to-noise ratio [SNR], contrast-to-noise ratio [CNR]) and qualitative evaluations of visibility and noise. ASPECTS was calculated for each reconstruction method, and diagnostic performance metrics, including sensitivity, specificity, and accuracy, were analyzed using DWI ASPECTS as the reference.

RESULTS

DLIR demonstrated significantly reduced image noise and superior SNR compared to FBP and IR (p < 0.0001). The CNR of DLIR was significantly higher than that of FBP (p = 0.0036), but no significant difference was observed compared to IR (p = 0.4238). Qualitative assessments indicated improved visibility and noise with DLIR. However, there were no significant differences in ASPECTS scores or the supplementary diagnostic performance metrics (sensitivity, specificity, accuracy) in predicting acute MCA territory infarction between DLIR and FBP, and between DLIR and IR.

CONCLUSION

DLIR improved image quality in thin-slice CT for acute MCA territory infarction but did not lead to significant differences in ASPECTS evaluation or the detection of EIC compared to FBP and IR.