Deep Learning-Based ASPECTS Algorithm Enhances Reader Performance and Reduces Interpretation Time.

Area of Science:

• Neuroradiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• The ASPECTS score is crucial for acute ischemic stroke treatment selection.
• Interpreting ASPECTS is challenging, time-consuming, and prone to interobserver variability.

Purpose of the Study:

• To evaluate the impact of a deep learning (DL) algorithm on ASPECTS interpretation accuracy and time.
• To assess the effect of DL assistance on interobserver reliability in ASPECTS scoring.

Main Methods:

• A multireader, multicase study involving 200 NCCT scans.
• Clinicians assessed ASPECTS with and without a DL-based algorithm (CINA-ASPECTS).
• Comparison of performance metrics (accuracy, ROC AUC) and interpretation time between assisted and unassisted assessments.

Main Results:

• Software assistance significantly improved region-based accuracy (72.4% to 76.5%) and ROC AUC (0.749 to 0.788).
• Interobserver reliability and correlation coefficient for ASPECTS evaluation improved significantly with DL aid.
• Mean interpretation time per case was reduced by 6% with the algorithm.

Conclusions:

• The DL algorithm enhances ASPECTS interpretation accuracy, consistency, and speed.
• This tool can improve patient selection for acute ischemic stroke treatment.
• The algorithm offers a potential solution to reduce variability and increase precision in ASPECTS evaluation.