Assessing accuracy and consistency in intracranial aneurysm sizing: human expertise vs. artificial intelligence

Area of Science:

• Vascular Neurology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Intracranial aneurysms (IAs) pose a significant rupture risk, with growth over 1 mm indicating increased danger.
• Accurate measurement of IAs is crucial for risk assessment and patient management.
• Current measurement practices involve human interpretation, which can introduce variability.

Purpose of the Study:

• To analyze the variability in intracranial aneurysm (IA) measurements performed by human raters and an Artificial Intelligence (AI) system.
• To compare the performance of junior and senior raters against an AI system in assessing IA dimensions.
• To evaluate the potential of AI in improving the accuracy and reliability of IA size assessments.

Main Methods:

• Utilized a dataset of 358 intracranial aneurysms (IAs) from 248 computed tomography angiography (CTA) scans.
• Employed four junior raters and one senior rater for manual measurements.
• Compared human measurements with assessments from an Artificial Intelligence (AI) system.

Main Results:

• Substantial inter- and intra-rater variability was observed among junior human raters.
• The senior rater exhibited lower intra-rater variability compared to junior raters.
• All human raters demonstrated standard deviations exceeding the 1 mm threshold for IA growth.
• A systemic bias showed human raters measuring IAs smaller than the AI system.
• The AI system achieved performance comparable to senior raters in monitoring IA growth.

Conclusions:

• Human assessment of intracranial aneurysms (IAs) presents significant variability challenges.
• Artificial Intelligence (AI) technology offers enhanced precision and reliability for IA measurements.
• AI shows particular promise in assisting junior raters and monitoring IA progression over time.