Development of an MRI based artificial intelligence model for the identification of underlying atrial fibrillation...

Atrial fibrillation (AF) represents a major risk factor of ischemic stroke recurrence with serious management implications. However, it often remains undiagnosed due to lack of standard or prolonged cardiac rhythm monitoring. We aim to create a novel end-to-end artificial intelligence (AI) model that uses MRI data to rapidly identify high AF risk in patients who suffer from an acute ischemic stroke.

This study comprises an internal retrospective cohort and a prospective cohort from Shanghai sixth people's hospital to train and validate an MRI-based AI model. Between January 1, 2018 and December 31, 2021, 510 patients were retrospectively enrolled for algorithm development and performance was measured using fivefold cross-validation. Patients from this trial were registered with http://www.chictr.org.cn, ChiCTR2200056385. Between September 1, 2022 and July 31, 2023, 73 patients were prospectively enrolled for algorithm test. An external cohort of 175 patients from Huashan Hospital, Minhang Hospital, and Shanghai Tenth People's Hospital was also enrolled retrospectively for further model validation. A combined classifier leveraging pre-defined radiomics features and de novo features extracted by convolutional neural network (CNN) was proposed to identify underlying AF in acute ischemic stroke patients. Area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were calculated for model evaluation.

The top-performing combined classifier achieved an AUC of 0.94 (95% CI, 0.90-0.98) in the internal retrospective validation group, 0.85 (95% CI, 0.79-0.91) in the external validation group, and 0.87 (95% CI, 0.90-0.98) in the prospective test group. Based on subgroup analysis, the AI model performed well in female patients, patients with NIHSS > 4 or CHA₂DS₂-VASc ≤ 3, with the AUC of 0.91, 0.94, and 0.90, respectively. More importantly, our proposed model identified all the AF patients that were diagnosed with Holter monitoring during index stroke admission.

Our work suggested a potential association between brain ischemic lesion pattern on MR images and underlying AF. Furthermore, with additional validation, the AI model we developed may serve as a rapid screening tool for AF in clinical practice of stroke units.

This work was supported by grants from the National Natural Science Foundation of China (NSFC, Grant Number: 81871102 and 82172068); Shanghai Jiao Tong University School of Medicine, Two-Hundred Talent Program as Research Doctor (Grant Number: SBR202204); Municipal Science and Technology Commission Medical Innovation Project of Shanghai, (Grant/Award Number: 20Y11910200); Research Physician Program of Shanghai Shen Kang Hospital Development Center (Grant Number: SHD2022CRD039) to Dr. Dong Huang and the SJTU Trans-med Awards Research (No. 20220101) to Dahong Qian.