Automated detection of new cerebral infarctions and prognostic implications using deep learning on serial MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Stroke Medicine

Background:

• Manual interpretation of follow-up MRI scans for stroke patients is time-consuming and prone to errors.
• Silent brain infarctions (SBIs) are often overlooked despite their significant prognostic implications for future stroke risk.

Purpose of the Study:

• To develop and validate a deep learning model for automated detection of new ischemic lesions on serial FLAIR MRI scans.
• To assess the clinical relevance of AI-detected SBIs in predicting subsequent stroke events.

Main Methods:

• A convolutional neural network was trained using supervised contrastive learning on 25,451 MRI slices from 1055 stroke patients.
• The model's performance was evaluated using internal and external validation cohorts, achieving an Area Under the ROC Curve (AUC) of 0.89.
• Clinical utility was assessed by analyzing an independent cohort for stroke recurrence risk in patients classified as SBI-positive by the model.

Main Results:

• The deep learning model demonstrated high accuracy in detecting new ischemic lesions.
• Patients identified as SBI-positive by the AI model exhibited a significantly increased risk of symptomatic stroke.
• In multivariable analysis, AI-detected SBIs were independently associated with a 3.8-fold higher risk of stroke recurrence.

Conclusions:

• Artificial intelligence can effectively identify clinically significant silent brain infarctions that are often missed in routine practice.
• Automated lesion detection using AI offers a reproducible imaging biomarker for risk stratification in stroke patients.
• This technology supports standardized MRI interpretation and can inform secondary stroke prevention strategies.