Prediction of Spontaneous Breathing Trial Outcome in Critically Ill-Ventilated Patients Using Deep Learning:...

Area of Science:

• Artificial Intelligence in Medicine
• Deep Learning for Healthcare
• Respiratory Therapy Innovations

Background:

• Long-term ventilator use is linked to poor quality of life, higher mortality, and increased costs.
• Manual ventilator weaning assessments are time-consuming, subjective, and prone to errors.
• Efficient and objective assessment tools are crucial for optimizing patient care.

Purpose of the Study:

• Develop an AI-based model to predict spontaneous breathing trial (SBT) success using pre-SBT clinical data.
• Provide an objective and efficient assessment tool for ventilator weaning.
• Enhance accuracy, reduce unnecessary SBTs, and optimize ICU resource utilization.

Main Methods:

• Retrospective cohort study involving 3686 adult ICU patients.
• Developed a novel hybrid CNN-MLP deep learning architecture for feature learning and fusion.
• Utilized 6536 pre-SBT clinical records for model training and validation.

Main Results:

• The AI model demonstrated high predictive performance.
• Achieved a precision of 99.3% and F1-score of 0.963 in the training dataset.
• Maintained strong accuracy in the test dataset with an 89.2% precision and 0.875 F1-score.

Conclusions:

• A deep learning-based model for SBT prediction was successfully developed.
• The model serves as an objective and efficient tool for ventilator weaning assessment.
• Integration into clinical workflows can improve patient care and reduce ventilator dependence.