Data-Driven Prognostication in Distal Medium Vessel Occlusions Using Explainable Machine Learning.

Area of Science:

• Neurology
• Artificial Intelligence in Medicine
• Cardiovascular Research

Background:

• Distal medium vessel occlusions (DMVOs) account for a significant proportion of acute ischemic stroke cases (25%-40%).
• Existing prognostic models are insufficient for DMVO-specific outcome prediction.
• Accurate prognostication is crucial for patient counseling and guiding research efforts.

Purpose of the Study:

• To develop and validate a machine learning model for predicting 90-day unfavorable outcomes in patients with primary DMVO.
• To identify key predictors of outcome in DMVO stroke patients.
• To create a tool for individualized prognostic predictions.

Main Methods:

• Retrospective analysis of 164 patients with primary DMVO.
• Development of a machine learning model using the TabPFN algorithm.
• Feature selection via least absolute shrinkage and selection operator (LASSO).
• Performance evaluation using 5-repeat 5-fold cross-validation, assessing discrimination and calibration.
• Identification of influential features using SHapley Additive Explanations (SHAP).

Main Results:

• The model achieved an area under the receiver operating characteristic curve (AUC) of 0.815, indicating good discrimination.
• A Brier score of 0.19 demonstrated good calibration of the model's predictions.
• Key predictors included admission NIHSS score, premorbid mRS, thrombectomy type, modified TICI score, and malignancy history.
• A deployed Web application facilitates individualized prognostication.

Conclusions:

• The developed machine learning model effectively predicts 90-day unfavorable outcomes in DMVO strokes with good accuracy.
• This prognostic tool supports personalized patient counseling and research in stroke care.
• The findings highlight the potential of AI in advancing precision medicine for stroke recovery.