A Novel Hybrid Ordinal Learning Model with Health Care Application.

Area of Science:

• Machine Learning
• Health Informatics
• Ordinal Learning

Background:

• Ordinal learning (OL) models are valuable in healthcare for tasks like disease grading and progression prediction.
• Training OL models is challenging when precise labels are scarce, but imprecise interval labels are abundant.
• Existing research on OL with imprecise/interval labels is limited.

Purpose of the Study:

• To develop a novel method for ordinal learning that integrates both precise and interval-labeled data.
• To address the common healthcare data challenge of limited precisely labeled samples.
• To improve the robustness and accuracy of ordinal learning models in practical applications.

Main Methods:

• Propose a Hybrid Ordinal Learner (HOL) model capable of handling mixed precise and interval labels.
• Develop an efficient optimization algorithm for the HOL formulation.
• Evaluate HOL against state-of-the-art OL methods on benchmark datasets.

Main Results:

• HOL significantly outperforms existing ordinal learning methods on four benchmarking datasets.
• The proposed HOL model achieves high accuracy in predicting the speed of progression to Alzheimer's Disease (AD) using multi-modal data.
• HOL demonstrates superior performance compared to existing methods in real-world AD progression prediction.

Conclusions:

• The Hybrid Ordinal Learner (HOL) offers a robust solution for ordinal learning with limited precise labels and abundant interval labels.
• HOL shows significant potential for advancing healthcare applications, particularly in predicting disease progression like that of Alzheimer's Disease.
• Accurate prediction of AD progression speed can aid in developing personalized intervention strategies for individuals with Mild Cognitive Impairment (MCI).