Mortality Prediction from Patient's First Day PAAC Radiograph in Internal Medicine Intensive Care Unit Using...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Informatics

Background:

• Predicting mortality in intensive care units (ICUs) is crucial for resource allocation and treatment planning.
• Chest radiographs are routinely obtained upon ICU admission.
• Early and accurate mortality prediction can significantly impact patient management strategies.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting mortality using initial chest radiographs.
• To identify key radiomic features predictive of mortality in ICU patients.
• To assess the performance of machine learning algorithms in mortality prediction.

Main Methods:

• Retrospective analysis of 510 ICU patients' chest radiographs.
• Data augmentation to increase dataset size to 3019 images for training/validation.
• Extraction and analysis of 74 radiomic features using machine learning algorithms.
• Evaluation using Area Under the ROC Curve (AUC), sensitivity, and specificity.

Main Results:

• Feature selection reduced 74 initial features to 10.
• The Subspace KNN algorithm achieved the highest prediction accuracy.
• Achieved an AUC of 0.88, sensitivity of 0.80, and specificity of 0.87.

Conclusions:

• Machine learning algorithms and radiomic features from chest radiographs are effective for ICU mortality prediction.
• Specific features like GLCM Contrast, Kurtosis, and Cardiomegaly are significant predictors.
• Integration into clinical decision support systems can enhance patient management.