Supporting diagnostic decisions using hybrid and complementary data mining applications: a pilot study in the...

Area of Science:

• Computational Medicine
• Artificial Intelligence in Healthcare
• Pediatric Emergency Medicine

Background:

• Pediatric emergency departments (EDs) face diagnostic challenges.
• Data mining (DM) offers potential for improved diagnostic support.
• A novel combination of DM methods was explored for pediatric diagnosis.

Purpose of the Study:

• To develop and evaluate a computer-based diagnostic system for pediatric emergency patients.
• To assess the accuracy of a combined data mining approach in diagnosis.
• To determine the utility of DM methods in supporting clinical decisions in the ED.

Main Methods:

• Simultaneous application of support vector machine (SVM), artificial neural networks (ANNs), fuzzy logic, and a voting algorithm.
• Utilized anonymized data from 26 clinical and laboratory parameters for each patient.
• The system was designed to classify patients into 18 distinct diagnoses.

Main Results:

• Retrospective analysis showed 98% diagnostic accuracy with the combined DM methods.
• Highest accuracy achieved for appendicitis (97%) and idiopathic thrombocytopenic purpura/erythroblastopenia (100%).
• Prospective testing resulted in 81% correct diagnoses by the system.

Conclusions:

• The combined DM methods effectively supported diagnosis using clinical and laboratory data.
• An optimized combination of complementary DM methods can assist medical decision-making in pediatric EDs.
• This approach demonstrates the potential of AI in enhancing diagnostic capabilities for pediatric emergency care.