Deep Learning Radiomics Based on Preoperational Ultrasound Images for Predicting Ipsilateral Ischemic Stroke in...

This study aimed to develop and validate an integrated model incorporating clinical, radiomic, and deep learning features to predict long-term ipsilateral ischemic stroke after carotid artery stenting (CAS).

We analyzed 802 patients who underwent CAS at three centers between 2018 and 2024. Clinical and ultrasound data were collected, and radiomic and deep learning features were extracted from preoperative plaque images. A combined model was built using Cox regression and random survival forest models, and then presented as a nomogram. Model performance was assessed using the C-index and Kaplan-Meier analysis.

Over a median follow-up of 62 months, 213 patients (26.6%) experienced ipsilateral stroke. We integrated the significant clinical, radiomics, and deep learning features (p<0.05) into a nomogram. The model demonstrated strong discriminative ability for stroke, with C-indices of 0.800 (95% CI: 0.759-0.841) in the training set, 0.751 (95% CI: 0.677-0.828) in the internal validation set, and 0.708 (95% CI: 0.634-0.782) in the external validation set. It significantly outperformed the conventional ultrasound model across all datasets (p<0.05). Moreover, Kaplan-Meier analysis confirmed that patients stratified into the high-risk group based on the nomogram had a substantially higher probability of ipsilateral stroke compared to the low-risk group (p<0.05).

A deep learning radiomics model based on preoperative ultrasound can predict long-time risk of ipsilateral stroke for patients with CAS, which may help in risk stratification and guide treatment decision-making and follow-up management.