Artificial intelligence assisted detection of large vessel occlusion on CT angiography in acute stroke patients: a multi-reader multi-case study
Nagaratnam K., Mathieson P., Podlasek A., Slade P., Ford GA., Cox A., Briggs JH., Woodhead ZVJ., Harston G.
Abstract Objectives We assessed the impact of artificial intelligence (AI) software (e-CTA, Brainomix) on clinical decision-making in patients with suspected acute ischemic stroke. Methods A retrospective, multi-reader-multi-case crossover design compared readers’ performance with vs without software support. Twenty cases were included, 10 with large vessel occlusion (LVO) and 10 without LVO. Twenty-one NHS clinicians, representing intended software users ranging in experience, conducted 2 sessions (washout period > 2 weeks). In session 1, software support was provided for 10 randomly selected cases. In session 2, support allocation was reversed. Outcome measures included LVO detection, collateral scoring, diagnosis, treatment decision, time taken and confidence. Results Sensitivity, specificity, and accuracy of LVO detection improved with imaging software for LVO detection, with increased confidence and reduced time taken. There was no significant difference in collateral scoring or diagnoses. Conclusion e-CTA can improve performance of NHS clinicians when interpreting acute stroke imaging. Advances in knowledge This paper provides new evidence that AI decision support software has the capacity to improve the performance of representative users in the NHS when interpreting imaging to identify patients for acute stroke treatments.