CTA-based deep-learning integrated model for identifying irregular shape and aneurysm size of unruptured intracranial aneurysms.

Abstract

Background: Artificial intelligence can help to identify irregular shapes and sizes, crucial for managing unruptured intracranial aneurysms (UIAs). However, existing artificial intelligence tools lack reliable classification of UIA shape irregularity and validation against gold-standard three-dimensional rotational angiography (3DRA). This study aimed to develop and validate a deep-learning model using computed tomography angiography (CTA) for classifying irregular shapes and measuring UIA size.

Methods: CTA and 3DRA of UIA patients from a referral hospital were included as a derivation set, with images from multiple medical centers as an external test set. Senior investigators manually measured irregular shape and aneurysm size on 3DRA as the ground truth. Convolutional neural network (CNN) models were employed to develop the CTA-based model for irregular shape classification and size measurement. Model performance for UIA size and irregular shape classification was evaluated by intraclass correlation coefficient (ICC) and area under the curve (AUC), respectively. Junior clinicians' performance in irregular shape classification was compared before and after using the model.

Results: The derivation set included CTA images from 307 patients with 365 UIAs. The test set included 305 patients with 350 UIAs. The AUC for irregular shape classification of this model in the test set was 0.87, and the ICC of aneurysm size measurement was 0.92, compared with 3DRA. With the model's help, junior clinicians' performance for irregular shape classification was significantly improved (AUC 0.86 before vs 0.97 after, P<0.001).

Conclusion: This study provided a deep-learning model based on CTA for irregular shape classification and size measurement of UIAs with high accuracy and external validity. The model can be used to improve reader performance.