Ischemic stroke clinical outcome prediction based on image signature selection from multimodality data

Area of Science:

• Medical imaging analysis
• Precision medicine
• Biostatistics

Background:

• Quantitative models are crucial for predicting health status in precision medicine.
• Current radiomics models often use excessive, redundant image features, neglecting individual feature importance for clinical outcome prediction.

Purpose of the Study:

• To develop a prognostic discrimination ranking strategy for selecting the most relevant image features in image-assisted clinical outcome prediction.
• To improve the accuracy and efficiency of radiomics models for predicting health status and preventing disease.

Main Methods:

• A novel redundancy and prognostic discrimination evaluation method to rank image features.
• Forward sequential feature selection to identify top-ranked relevant features.
• Fusion of selected image features with clinical parameters for classification model input.

Main Results:

• The proposed model demonstrated improved performance over five other feature selection models.
• Achieved high Area Under the Curve (AUC) values (0.821 to 1) in predicting five clinical outcome scores.
• Successfully trained and tested on 70 patient studies with six MR sequences and four clinical parameters.

Conclusions:

• The developed prognostic discrimination ranking strategy effectively selects relevant image features for clinical outcome prediction.
• This approach enhances the performance of radiomics models in precision medicine.
• The fusion of selected image features and clinical parameters leads to more accurate disease prediction and prevention.