Artificial intelligence-driven translational medicine: a machine learning framework for predicting disease outcomes...

Area of Science:

• Medical Informatics
• Computational Biology
• Translational Medicine

Background:

• Artificial intelligence (AI) and machine learning (ML) are transforming medicine, enabling better disease prediction and patient care.
• Challenges like data heterogeneity and scalability hinder optimal predictive performance in current AI/ML models.
• Addressing these limitations is crucial for advancing personalized medicine and clinical decision support.

Purpose of the Study:

• To propose a novel AI-based framework integrating Gradient Boosting Machines (GBM) and Deep Neural Networks (DNN).
• To evaluate the framework's performance on diverse datasets, including MIMIC-IV and UK Biobank.
• To overcome challenges in data heterogeneity, class imbalance, and scalability for improved predictive modeling.

Main Methods:

• Developed a hybrid AI framework combining GBM and DNN.
• Validated the framework using the MIMIC-IV critical care database and the UK Biobank dataset.
• Assessed performance using Accuracy, Precision, Recall, F1-Score, and AUROC against established ML models.

Main Results:

• The proposed framework significantly outperformed classical models like Logistic Regression, Random Forest, SVM, and standard Neural Networks.
• Achieved an AUROC of 0.96 on the UK Biobank dataset, surpassing Neural Networks (0.92).
• Demonstrated high efficiency with rapid training times (32.4s on MIMIC-IV) and low prediction latency.

Conclusions:

• The AI framework effectively addresses key challenges in translational medicine, offering superior predictive accuracy and efficiency.
• Its robust performance across varied datasets indicates strong potential for real-time clinical decision support systems.
• Future work will focus on enhancing scalability and interpretability for broader clinical adoption and improved patient outcomes.