Akademik Veri Yönetim Sistemi
Microchemical Journal, cilt.218, 2025 (SCI-Expanded)
Bladder cancer (BCa) is a common malignancy with high recurrence rates, highlighting the need for non-invasive, objective monitoring tools. Current methods, such as cystoscopy and histopathology, are invasive, subjective, and often lack sensitivity for early or recurrent disease. This study investigates Explainable Artificial Intelligence (XAI)-assisted Attenuated Total Reflection-Fourier Transform Infrared (ATR-FTIR) spectroscopy as a rapid and transparent approach for tracking BCa progression and recurrence. Urine samples (n = 80) from benign, primary, relapse, and healthy control patients were analyzed, yielding 3351 spectral features. Seven machine learning (ML) models were trained and assessed using stratified five-fold cross-validation. To enhance interpretability, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were applied to identify spectral biomarker bands linked to BCa stages. Random Forest emerged as the optimal model, achieving the highest balance between accuracy and stability (mean ROC AUC: 0.798 ± 0.041). XAI analyses of this model highlighted key spectral biomarkers at 3997, 3937, 3645, and 2071 cm−1 as influential for distinguishing BCa subtypes. SHAP revealed global feature contributions, while LIME provided patient-level spectral insights. XAI-integrated ATR-FTIR spectroscopy shows promise as a transparent, non-invasive, and data-driven strategy for BCa monitoring. By enhancing interpretability and identifying stage-specific spectral biomarkers, this approach may support personalized disease management. Further validation in larger patient cohorts is warranted to establish clinical applicability.