Multiple Cost Optimisation for Alzheimer's Disease Diagnosis.

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Machine Learning

Background:

• Current machine learning models for dementia diagnosis often overlook practical constraints such as assessment time and financial costs.
• Optimizing dementia care pathways requires addressing multiple, non-fungible budget limitations faced by stakeholders.

Purpose of the Study:

• To develop a generalized cost-sensitive feature selection approach for dementia diagnosis that accounts for multiple budget constraints.
• To introduce an optimization algorithm capable of searching cost-weighting hyperparameters under total budget limitations.

Main Methods:

• Developed a novel optimization algorithm for cost-sensitive feature selection with multiple budget constraints.
• Implemented and validated the framework using assessment time and financial budget costs.
• Created a user-friendly graphical user interface (GUI) sandbox tool for a humans-in-the-loop approach.

Main Results:

• Budget constraints effectively controlled the feature selection process in a practical manner.
• Adjusting cost-weighting hyperparameters expanded the range of selected features.
• The GUI tool facilitated adoption and exploration by non-technical users.

Conclusions:

• Setting initial budget constraints and fine-tuning cost-weighting hyperparameters is an effective strategy for feature selection with multiple cost considerations.
• This approach enables more realistic optimization and redesign of dementia diagnostic assessments.
• The framework supports the development of predictive yet practical dementia diagnostic tools suitable for clinical implementation.