Integrated Deep Learning Model for the Detection, Segmentation, and Morphologic Analysis of Intracranial Aneurysms...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Unruptured intracranial aneurysms (UIAs) require precise morphologic measurement for risk assessment.
• Current measurement methods can be subjective and time-consuming.
• Deep learning offers potential for automated and accurate analysis of medical imaging data.

Purpose of the Study:

• To develop and validate an integrated deep learning (IDL) model for UIA detection, segmentation, and morphologic measurement using CT angiography (CTA) data.
• To assess the IDL model's performance on a multicenter external testing set.
• To evaluate the IDL model's ability to assist junior radiologists in UIA morphologic analysis.

Main Methods:

• A retrospective study utilizing CTA data from a tertiary referral hospital for training and a multicenter dataset for external validation.
• Development of an IDL model based on the nnU-Net algorithm for UIA detection, segmentation, and morphologic measurement.
• Performance evaluation using Dice Similarity Coefficient (DSC) and Intraclass Correlation Coefficient (ICC), with senior radiologist measurements as the reference standard.

Main Results:

• The IDL model achieved 97% accuracy in UIA detection and a DSC of 0.90 for segmentation.
• Model-based morphologic measurements demonstrated good agreement with reference standards (ICCs > 0.85 in training, > 0.80 in external testing).
• Junior radiologists assisted by the IDL model showed significantly improved UIA size measurement accuracy (ICC improved from 0.88 to 0.96, P < .001).

Conclusions:

• The developed IDL model shows high accuracy in detecting, segmenting, and measuring UIAs from CTA data.
• The IDL model performs reliably on multicenter external validation datasets.
• This AI tool can effectively assist less experienced radiologists in the morphologic analysis of UIAs, potentially improving diagnostic consistency.