Deep learning-based automatic ASPECTS calculation can improve diagnosis efficiency in patients with acute ischemic...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• The Alberta Stroke Program Early CT Score (ASPECTS) is crucial for evaluating ischemic changes in acute ischemic stroke using non-contrast computed tomography (NCCT).
• Current ASPECTS interpretation relies heavily on expert experience, leading to inter-reader variability and potential delays in treatment.
• There is a need for a standardized, objective, and efficient method for ASPECTS assessment in clinical practice.

Purpose of the Study:

• To develop and validate a deep learning (DL) based system for automated ASPECTS assessment in acute ischemic stroke.
• To evaluate the clinical applicability and efficiency of the automated DL system compared to expert interpretations.

Main Methods:

• A DL system was trained and validated on 1987 NCCT scans from four centers.
• The system's performance was assessed against physician consensus (reference standard) using metrics like AUC and ICC.
• Prospective validation was conducted on 13,399 patients in real-world clinical settings.

Main Results:

• The DL system achieved an AUC of 84.97% and an ICC of 0.84 on the test cohort.
• Diagnostic sensitivity was 94.61% for ASPECTS ≥ 6, with substantial agreement (ICC=0.65) with experts.
• The system reduced diagnosis time by 74.8% (from 130.6s to 33.3s) and was utilized in 94.0% of prospective cases, with 96% of physicians noting improved efficiency.

Conclusions:

• The developed DL-based system accurately and rapidly determines ASPECTS, potentially streamlining clinical workflows for early stroke intervention.
• The automated system demonstrates non-inferiority to expert ratings, improving evaluation consistency and significantly reducing processing time.
• While effective, physician validation remains essential for clinical application of the automated ASPECTS system.