The analysis and classification of electronic health records (EHR) for effective patient diagnosis remain a critical challenge in healthcare, particularly when dealing with high dimensional imbalanced medical datasets. In this presentation, I will introduce MedMGF, a novel multi-level graph-based framework designed to improve classification performance by integrating patient medical profiles and their relationships in a unified architecture. The presentation will be divided into two key parts:

Patient Medical Profile Representation: We develop a comprehensive approach to model multi-layered patient medical profiles extracted from hospitalization records, embedding critical health data in a graph-based format. These profiles are used to construct a patient-patient network, where connections are determined by similarities in medical histories.

Classification on Imbalanced Data: To tackle the inherent class imbalance in medical datasets, we propose a modification to the Focal Loss (FL) function. This approach, combined with the MedMGF architecture, enables a better classification performance compared to traditional Graphical Convolutional Networks with Binary Cross Entropy, FL, and Synthetic Minority Oversampling Technique (SMOTE).

This framework holds potential for various healthcare applications, particularly in improving early diagnosis and patient management in critical conditions like pediatric sepsis.

Nguyen Tuong Minh is a last year PhD student at Department of Industrial Systems Engineering and Management at the National University of Singapore, advised by Prof. Poh Kim Leng, Dr. Lee Jan Hau and Dr. Chong Shu-Ling from KK Women’s and Children’s Hospital, Singapore. She received Bachelor’s degree in Computer Science at University of Science, National University of Vietnam, and Master of Science in the Department of Industrial Systems Engineering and Management at the National University of Singapore. Her research interests cover disease diagnosis modelling as well as applied machine learning in healthcare.