Related Experiment Video
Updated: Oct 5, 2025
04:57
Author Spotlight: Minimally Invasive Ultrasound-Guided Acupotomy in Knee Osteoarthritis Treatment
Published on: April 26, 2024
739
Explainable machine learning for knee osteoarthritis diagnosis based on a novel fuzzy feature selection methodology
Christos Kokkotis1,2, Charis Ntakolia3,4, Serafeim Moustakidis5
1Institute for Bio-Economy & Agri-Technology, Center for Research and Technology Hellas, 38333, Volos, Greece. chkokkotis@gmail.com.
Physical and Engineering Sciences in Medicine
|January 31, 2022
Summary
This study introduces a new fuzzy ensemble feature selection method to identify key risk factors for knee osteoarthritis (KOA). The approach achieved 73.55% accuracy, improving KOA diagnosis.
Area of Science:
- Biomedical data science
- Computational biology
- Medical informatics
Background:
- Knee Osteoarthritis (KOA) is a degenerative joint disease characterized by progressive cartilage loss.
- The multifactorial nature and complex pathophysiology of KOA contribute to diagnostic challenges and errors.
- High dimensionality and data heterogeneity in public datasets like the Osteoarthritis Initiative (OAI) complicate KOA risk factor identification.
Purpose of the Study:
- To develop a robust Feature Selection (FS) methodology for handling multidimensional KOA data.
- To improve the identification of critical risk factors for KOA diagnosis by overcoming limitations of existing FS techniques.
- To enhance the accuracy and reliability of clinical KOA diagnosis through advanced analytics.
Main Methods:
- Utilized multidimensional data from the Osteoarthritis Initiative (OAI) database.
- Developed a fuzzy ensemble feature selection methodology integrating multiple FS algorithms (filter, wrapper, embedded) using fuzzy logic.
- Evaluated the methodology through an extensive experimental setup comparing against competing FS algorithms and machine learning (ML) models.
Main Results:
- The proposed fuzzy ensemble FS methodology effectively handled high-dimensional KOA data.
- The best performing model, a Random Forest classifier, achieved 73.55% classification accuracy using twenty-one selected risk factors.
- Explainability analysis was conducted to understand the impact of selected features on the model's predictions.
Conclusions:
- The developed fuzzy ensemble FS methodology offers a robust approach for identifying significant risk factors in KOA.
- The study demonstrates the potential of advanced analytics and feature selection in improving KOA diagnosis accuracy.
- Enhanced understanding of KOA's underlying mechanisms is facilitated by explainable AI techniques applied to selected risk factors.
