MRI radiomic features-based machine learning approach to classify ischemic stroke onset time.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Accurate determination of time since stroke onset (TSS) is crucial for effective acute ischemic stroke management.
• Current methods for TSS estimation can be imprecise, especially in cases of unwitnessed strokes.
• Advanced computational techniques offer potential for improved TSS classification.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models utilizing MRI radiomics features for classifying time since stroke onset (TSS).
• To develop and validate models capable of aiding in stroke assessment and guiding treatment decisions, particularly for thrombolysis.

Main Methods:

• Eighty-four patients with anterior circulation ischemic stroke were included, split into training (n=51) and independent test (n=33) cohorts.
• Radiomic features were extracted from segmented infarct regions using 3D-slicer and processed in R, yielding 4312 features per sequence.
• Six ML models were trained to classify TSS, with performance evaluated using ROC curves and other metrics.

Main Results:

• A deep learning model employing DWI/ADC radiomic features demonstrated the best performance for binary TSS classification (≤4.5 h) in the independent test cohort.
• This model achieved an Area Under the Curve (AUC) of 0.754, with an accuracy of 0.788.
• Incorporating clinical data did not enhance the performance of the DWI/ADC-based deep learning model.

Conclusions:

• A novel deep learning model utilizing DWI/ADC radiomic features effectively classifies time since stroke onset.
• This model shows promise in assisting clinical decision-making for thrombolysis in patients with unknown stroke onset times.
• The developed TSS prediction models are accessible for further research and application.