LTPat-based SOXFE: A self-organized explainable feature engineering model for automated fibromyalgia detection
Technology and Health Care, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1177/09287329261495183
- Dergi Adı: Technology and Health Care
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, CINAHL, EMBASE, INSPEC, MEDLINE, Academic Search Ultimate (EBSCO), Biomedical Reference Collection: Corporate Edition (EBSCO), Engineering Source (EBSCO), Health Research Premium Collection (ProQuest)
- Anahtar Kelimeler: EEG signal analysis, explainable artificial intelligence, feature engineering, Fibromyalgia detection, model-derived channel-transition diagram
- Ankara Yıldırım Beyazıt Üniversitesi Adresli: Evet
Özet
Background: Fibromyalgia is a chronic pain disorder associated with sleep disturbance and substantial functional burden. Objective: This study evaluated a self-organized explainable feature engineering (SOXFE) model for EEG-based fibromyalgia classification. Methods: A previously reported sleep EEG dataset from a 32-participant source cohort was analyzed separately for NREM Stage 2 and Stage 3. The analysis-ready version contained 136 labeled stage-record files: 74 for Stage 2 and 62 for Stage 3. The SOXFE model comprised LTPat feature extraction, training-only CWNCA feature selection, tkNN classification, IMV information fusion, and DLob-based explanation. Record-wise LORO was primary; segment-level 10-fold cross-validation was complementary. Results: Record-wise LORO used 74 outer folds for Stage 2 and 62 for Stage 3. Descriptive pooled held-out-epoch accuracy was 92.77% (95% CI: 91.76–93.67) for Stage 2 and 97.00% (95% CI: 96.22–97.63) for Stage 3. Every epoch from the held-out recording was excluded from training and selection. Complementary segment-level 10-fold cross-validation yielded 100.00% accuracy for both stages but was not used as evidence of record-wise generalization. Conclusions: The model provided record-disjoint classification and traceable model-derived explanations in this single-source dataset. LORO prevented within-record epoch leakage. However, participant linkage was unavailable, so records from the same person may have occurred across outer partitions and the reported values may overestimate person-independent performance. Verified subject-wise validation is required before clinical use.