Binary Classification Using Neural and Clinical Features: An Application in Fibromyalgia With Likelihood-Based...

Area of Science:

• Neuroscience
• Medical Informatics

Background:

• Clinical diagnosis often relies on behavioral performance, questionnaires, and physical exams.
• Integrating neuroimaging findings into diagnostic models is challenging due to data collection and integration difficulties.
• Current methods struggle to build large cohorts for disease feature description.

Purpose of the Study:

• To develop a robust framework for fibromyalgia detection by merging data from multiple clinical sites.
• To integrate functional near-infrared spectroscopy (fNIRS) data with self-reported clinical features for improved classification.
• To implement a decision-level fusion strategy for combining outputs from multiple classifiers.

Main Methods:

• A framework for fibromyalgia detection using likelihood-based decision-level fusion was implemented.
• Functional connectivity maps derived from fNIRS data across three tasks were used as features.
• Self-reported clinical features were incorporated alongside neuroimaging data.
• Five classifiers (kNN, LDA, SVM) were trained and their outputs fused.
• The framework tolerates missing classifier inputs due to data limitations.

Main Results:

• Fusion of classification opinions using likelihood ratios significantly outperformed individual classifiers.
• Achieved 100% sensitivity and specificity when fusing 2, 3, 4, or 5 different classifiers.
• The proposed method demonstrated robustness in handling missing data from certain classifiers.

Conclusions:

• The developed framework effectively integrates multi-site data and diverse features for enhanced diagnostic accuracy.
• Decision-level fusion of classifiers is a viable strategy for improving fibromyalgia detection using fNIRS and clinical data.
• The method shows potential for widespread clinical application due to its robustness and high performance.