Development of a Machine Learning Model for Rapid Blood Infection Diagnosis

Thesis Abstract

Bloodstream infections represent a significant global health challenge, contributing to high morbidity and mortality rates due to delayed diagnosis and initiation of targeted therapy. Traditional diagnostic methods such as blood culture, though considered the gold standard, often require 24 to 48 hours to yield results, thereby impeding timely clinical decision-making and patient management. This study aims to develop a robust machine learning-based diagnostic model capable of rapidly identifying blood infections directly from clinical laboratory data, thereby reducing diagnostic latency and improving patient outcomes. The primary objectives are to (1) compile and preprocess a comprehensive dataset comprising clinical, microbiological, and biochemical parameters from blood samples, (2) evaluate and compare the performance of various machine learning algorithms—including Random Forest, Support Vector Machines, and Gradient Boosting—in classifying blood infection status, (3) identify the most predictive features and assess their clinical relevance, and (4) validate the developed model using independent datasets to ensure generalizability and robustness. The research adopts a quantitative, cross-sectional design, incorporating data collected from 1,200 patients presenting with suspected bloodstream infections at a tertiary medical center over a 12-month period. Data collection involves extracting variables such as complete blood counts, C-reactive protein levels, procalcitonin, blood culture results, patient demographics, and clinical histories. These data are obtained from electronic health records and laboratory information systems, with strict adherence to ethical standards including patient confidentiality and data anonymization. The instruments used include standardized laboratory test results and structured data extraction forms validated through expert review. Data preprocessing includes handling missing values, normalization, feature encoding, and entropy-based feature selection to enhance model performance and interpretability. Analytical techniques encompass supervised machine learning algorithms trained using 80% of the dataset, with the remaining 20% allocated for testing and validation. Model performance is assessed through metrics such as accuracy, sensitivity, specificity, precision, F1-score, and area under the receiver operating characteristic curve. Feature importance analysis employs techniques like permutation importance and SHAP (SHapley Additive exPlanations) values to determine the most influential predictors. Model validation is further performed using k-fold cross-validation and external datasets obtained from neighboring hospitals to ensure reproducibility across different populations. Expected findings include the identification of a machine learning algorithm, likely Gradient Boosting, that achieves over 92% accuracy and an area under the ROC curve exceeding 0.95 in diagnosing blood infections rapidly. The study anticipates revealing critical biomarkers and laboratory parameters—such as procalcitonin levels and white blood cell counts—that substantially contribute to predictive accuracy. The findings are expected to demonstrate that the machine learning model outperforms traditional clinical assessments and laboratory screening methods in both speed and reliability, establishing a promising tool for real-time blood infection diagnosis. This research contributes novel insights into the application of advanced machine learning techniques in infectious disease diagnosis, particularly in resource-constrained healthcare settings, by providing an efficient, scalable, and interpretable model. It aligns with theoretical frameworks underpinning supervised learning and data-driven clinical decision support, notably the Health Belief Model and the Theory of Planned Behavior, to emphasize the integration of technological innovation into clinical practice. The study’s implications include facilitating early initiation of targeted antimicrobial therapy, reducing unnecessary antibiotic use, and ultimately decreasing infection-related mortality. In conclusion, this study provides a comprehensive model for rapid blood infection detection leveraging machine learning, with significant potential to transform diagnostic processes. It recommends further research into integrating such models into clinical workflows, exploring real-time data streams, and expanding to include multi-modal data such as imaging and genomic information to enhance diagnostic precision. The findings pave the way for future developments in AI-driven clinical decision support systems in infectious disease management.