Double decoupled network for imbalanced obstetric intelligent diagnosis.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Machine Learning

Background:

• Electronic Medical Records (EMRs) are crucial for intelligent diagnosis.
• EMR diagnostic results are multi-disease, imbalanced, and highly coupled.
• Traditional rebalancing methods fail on coupled imbalanced datasets.

Purpose of the Study:

• To propose a novel intelligent diagnosis model for EMRs.
• To address the challenges of imbalanced and coupled diagnostic data.
• To improve the accuracy of multi-label classification in EMRs.

Main Methods:

• Developed the Double Decoupled Network (DDN) model.
• Utilized Convolutional Neural Networks (CNN) for representation learning.
• Introduced the Decoupled and Rebalancing highly Imbalanced Labels (DRIL) algorithm for classification.

Main Results:

• DDN achieved high accuracy on Chinese Obstetric EMR (COEMR) datasets: 84.17%.
• Validated effectiveness on benchmark datasets: AAPD (86.35%) and RCV1 (93.87%).
• Outperformed current optimal experimental results on all tested datasets.

Conclusions:

• The DDN model effectively handles imbalanced and coupled diagnostic data in EMRs.
• DDN demonstrates effectiveness and universality across medical and general text classification tasks.
• The proposed DRIL algorithm successfully decouples and rebalances imbalanced labels.