Automated Calculation of Alberta Stroke Program Early CT Score: Validation in Patients With Large Hemispheric Infarct

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Large hemispheric infarcts require prompt assessment for effective treatment.
• Early detection of brain ischemia is crucial for stroke management.
• Computed tomography (CT) and diffusion-weighted imaging (DWI) are key imaging modalities.

Purpose of the Study:

• To compare the accuracy of machine learning-based automated software (RAPID ASPECTS) versus experienced clinicians in calculating CT ASPECTS.
• To evaluate the agreement of automated and clinician-based CT ASPECTS with DWI ASPECTS as a reference standard.
• To determine the efficacy of RAPID ASPECTS in identifying early signs of brain ischemia.

Main Methods:

• Retrospective analysis of CT and MRI scans from the GAMES-RP study involving patients with large hemispheric infarcts.
• Blinded evaluation of CT and DWI ASPECTS by experienced readers and automated software (RAPID ASPECTS).
• Comparison of CT ASPECTS (clinician median and automated) with DWI ASPECTS using interclass correlation coefficient.

Main Results:

• RAPID ASPECTS showed a higher agreement with DWI ASPECTS than the median CT ASPECTS from clinicians.
• Automated scoring demonstrated a significantly lower median error compared to clinician scoring (1 vs. 3, P<0.001).
• The automated score was more accurate in both full-scale and dichotomized ( <6 vs. ≥6) assessments.

Conclusions:

• RAPID ASPECTS is more accurate than experienced clinicians in identifying early brain ischemia on CT scans.
• Automated software offers a reliable tool for assessing stroke severity using ASPECTS.
• Machine learning enhances the diagnostic accuracy in acute stroke imaging.