A Multiscale Entropy Based Machine Learning Approach for Predicting Speech Recovery after Total Laryngopharyngectomy

This study aims to address the clinical challenge of variable outcomes in speech restoration after total laryngopharyngectomy (TLP) with J-shaped anterolateral thigh (ALT) flap reconstruction. To this end, this study proposes a multiscale entropy (MSE)-based machine learning (ML) framework to characterize flap biomechanics and predict postoperative speech recovery. A total of 26 TLP patients (19 with phonation and 7 with whisper) were recruited, and ultrasound radiofrequency data were obtained from longitudinal J-shaped ALT flap views. Envelope signals were extracted, and MSE features across scales 1 to 10 were computed. After feature selection with least absolute shrinkage and selection operator (LASSO) regression, three classifiers, including support vector machine (SVM), random forest (RF), and linear discriminant analysis (LDA), were trained under 30 repetitions of five-fold stratified cross-validation. The RF model achieved the best performance with an area under the receiver operating characteristic curve (AUROC) of 0.97, accuracy of 93.23%, and sensitivity of 94.83%. These findings highlight the potential of MSE-derived features, combined with ML, as noninvasive biomarkers for predicting speech recovery after TLP.