A multi-voting enhancement for newborn screening healthcare information system

Area of Science:

• Biochemistry
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Neonatal metabolic disorders often present with subtle clinical symptoms, risking severe, irreversible developmental damage or death if untreated.
• Early detection through newborn screening is vital for preventing long-term adverse outcomes in infants.
• Methylmalonic Acidemia (MMA) is one such critical metabolic disorder requiring accurate and timely diagnosis.

Purpose of the Study:

• To develop and validate an advanced computational model for the accurate screening of Methylmalonic Acidemia (MMA) in newborns.
• To enhance the predictive accuracy, sensitivity, and specificity of existing newborn screening methods for metabolic disorders.
• To create a versatile model applicable to the diagnosis of other neonatal metabolic diseases.

Main Methods:

• Implementation of a Support Vector Machine (SVM) based model incorporating Feature Selection techniques.
• Utilization of Grid Search and Cross-Validation for model optimization and robust performance evaluation.
• Integration of a multi-model Voting Mechanism to improve diagnostic decision-making.

Main Results:

• The developed SVM model demonstrated significantly improved predicting accuracy for MMA.
• Enhanced sensitivity and specificity were achieved in identifying infants with Methylmalonic Acidemia.
• The model's architecture proved effective in interpreting complex metabolic disorder data.

Conclusions:

• The proposed enhanced SVM model offers a powerful tool for improving newborn screening accuracy for MMA.
• This approach holds significant potential for early detection and prevention of severe neonatal complications.
• The model's adaptability suggests its utility in screening for a broader range of metabolic diseases in neonates.