A novel early diagnostic framework for chronic diseases with class imbalance

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Chronic diseases pose significant global health challenges due to complex clinical presentations and complications.
• Early diagnosis of chronic diseases is crucial for effective management and improved patient outcomes.
• Existing machine learning approaches struggle with feature hiding and imbalanced data in chronic disease datasets.

Purpose of the Study:

• To develop a universal and efficient diagnostic framework for timely and accurate early diagnosis of chronic diseases.
• To address the limitations of feature hiding and imbalanced class distribution in chronic disease datasets.
• To enhance the diagnostic performance for identifying sick cases in imbalanced datasets.

Main Methods:

• Proposed a network-limited polynomial neural network (NLPNN) algorithm to capture hidden high-level features and prevent overfitting.
• Developed an attention-empowered NLPNN algorithm to specifically improve diagnostic accuracy for minority classes in imbalanced datasets.
• Validated the framework on eleven chronic disease datasets, including public and real-world imbalanced data.

Main Results:

• The proposed NLPNN-based framework demonstrated superior performance compared to state-of-the-art machine learning algorithms.
• Achieved significant improvements in key performance metrics: accuracy, recall, F1-score, and G-mean.
• The attention-empowered NLPNN effectively enhanced the detection of chronic diseases, particularly in imbalanced datasets.

Conclusions:

• The developed diagnostic framework offers an efficient and accurate solution for the early detection of chronic diseases.
• The NLPNN and its attention-empowered variant successfully mitigate feature hiding and class imbalance issues.
• This framework has the potential to significantly aid clinicians in timely and accurate chronic disease diagnosis.