AI-driven prognostics in pediatric bone marrow transplantation: a CAD approach with Bayesian and PSO optimization

Area of Science:

• Medical Informatics
• Computational Biology
• Hematology

Background:

• Bone marrow transplantation (BMT) is crucial for treating pediatric hematological diseases but faces challenges in donor-recipient matching and complication prediction.
• Machine learning (ML) and artificial intelligence (AI) offer advanced analytical capabilities to address these complexities in BMT.

Purpose of the Study:

• To develop and evaluate a novel Computer-Aided Diagnosis (CAD) framework utilizing ML/AI for optimizing donor-recipient matching in pediatric allogeneic BMT.
• To enhance the prediction of BMT success rates and patient survival by analyzing critical factors like genetic compatibility and human leukocyte antigen types.

Main Methods:

• A novel CAD framework was developed, incorporating Particle Swarm Optimization for feature selection and an ensemble of seven ML models.
• Bayesian optimization using Adaptive Tree of Parzen Estimators (TPE) was employed for hyperparameter tuning, alongside L1/L2 normalization for data preprocessing.
• The Local Interpretable Model-Agnostic Explanations (LIME) framework was used to ensure model transparency and interpretability.

Main Results:

• The study analyzed a dataset titled "Bone Marrow Transplant: Children," identifying key factors influencing survival, including Donorage, extcGvHD, PLTrecovery, and survival_time.
• The optimal CAD framework achieved high performance metrics: 98.07% accuracy, 98.08% Balanced Accuracy, 98.45% precision, 98.02% recall, 98.14% specificity, 98.23% F1 score, and 96.53% Intersection over Union.
• Statistical validation using ANOVA and T-tests confirmed significant associations between identified factors and patient survival status.

Conclusions:

• The proposed AI-driven CAD framework significantly enhances donor-recipient matching and survival prediction in pediatric BMT.
• The framework provides interpretable insights, serving as a valuable clinical tool for improving outcomes in childhood bone marrow transplantation.
• The study highlights the potential of advanced ML/AI techniques to overcome challenges in complex medical procedures like BMT.