多重影像典範基兒童動態磁振造影語音道解剖分割

// Multi-Atlas-Based Segmentation of Pediatric Vocal Tract Anatomy in Dynamic Magnetic Resonance Imaging

本研究比較四種多重圖譜標籤融合方法，用於動態MRI中舌頭、軟顎和腺樣體解剖結構分割｜校正學習（CL）展現最高準確度，隨時間幀數增加，Dice相似係數由0.89提升至0.92｜結果顯示CL為兒童聲道分析中自動且可靠的最佳分割方式。

// This study compared four multi-atlas label fusion methods to segment tongue, velum, and adenoid anatomy in dynamic MRI during speech. Corrective learning (CL) showed the highest accuracy, improving significantly with more temporal frames, while other methods remained relatively flat. CL achieved an average dice similarity coefficient increase from 0.89 to 0.92 as frames increased from 10 to 40. The results suggest CL is the best approach for automatic, reliable segmentation in pediatric vocal tract analysis.

Published 2026年8月12日