摘要
Restoring speech after total laryngopharyngectomy (TLP) with flap reconstruction remains challenging. This study evaluated the feasibility of ultrasound scatteromics for assessing postoperative speech recovery by characterizing reconstructed flap tissues. Twenty-six patients were classified into phonation and whisper groups. Quantitative ultrasound (QUS) imaging based on the Nakagami parameter (m), homodyned K distribution parameters (alpha and k), and entropy (H) was performed from three anatomical views. For each parametric image, 13 first-order statistical features were extracted and selected to construct scatteromics features, which were used to train machine learning classifiers with repeated five-fold stratified cross-validation. Model performance and probabilistic reliability were evaluated using the area under the receiver operating characteristic curve (AUROC), calibration analysis, and the Brier score, and a simplified scatteromics model using four core parameters (m, alpha, k, and H) was assessed for comparison. Scatteromics derived from the reconstructed neopharynx view yielded consistently high discriminative performance across classifiers, with AUROC values ranging from 0.90 to 0.99. The simplified scatteromics approach demonstrated comparably strong classification performance, whereas individual QUS envelope statistics parameters showed limited discriminative ability (AUROC < 0.70). Probabilistic evaluation indicated that the full scatteromics framework yielded more stable probability estimation under limited sample size. Scatteromics may serve as a physically interpretable approach for characterizing reconstructed flap tissue and exploring speech recovery after TLP.