A questionnaire survey on impact of implementing a brain computed tomography deep learning model of intracranial haemorrhage for direct clinician decision support in emergency departments

Abstract Background Noncontrast computed tomography of the brain (NCCTB) has been widely used to detect intracranial haemorrhage (ICH). Deep learning models (DLMs) have been advocated as decision‐support tools to reduce diagnostic errors. A DLM for detecting ICH and midline shift in NCCTB was deployed in 17 public emergency departments (EDs) under the Hospital Authority in Hong Kong. This study aims to assess emergency physicians' knowledge, attitudes, confidence and trust regarding the use of this DLM for interpreting NCCTB. Methods A cross‐sectional survey was performed. All doctors working in EDs from March to June 2025 were invited to participate in an online survey. Subgroup analyses were performed to explore the relationships among background variables, knowledge, attitudes and confidence. Results The survey response rate was 42.4% ( n = 236). A total of 93.2% ( n = 220) of the participants agreed that DLM is trustworthy. A higher percentage of emergency physicians with satisfactory understanding of the diagnostic performance of the DLM ( n = 127, 53.8%) reported shorter documentation (53.5% vs. 32.1%, p < 0.001) and decision‐making times (65.3% vs. 55.9%, p = 0.003), higher diagnostic accuracy (92.1% vs. 87.2%, p = 0.006), greater confidence in discharging patients (92.1% vs. 87.1%, p < 0.001) and improved efficiency in NCCTB analysis (84.2% vs. 78.0%, p < 0.001) than those with an unsatisfactory knowledge level. However, experience in EDs did not predict frequency of use (higher frequency in 97.3% of trainees vs. 92.9% of specialists, p = 0.072), familiarity with the model (greater familiarity in 73.7% of trainees vs. 85% of specialists, p = 0.338), or time required for documentation (shorter time in 48.7% of trainees vs. 38.6% of specialists, p = 0.153). Conclusions Emergency physicians who understand the DLM better reported better trust, higher efficiency, diagnostic accuracy and shorter documentation and decision‐making times when using the DLM.