利用機器學習判斷兒童睡眠呼吸中止症之模型與系統

經由臉部特徵點的分析，有助兒童睡眠呼吸中止症的輔助判斷。傳統上，標定臉部特徵點都是採用人工判定的方式。在本篇論文，我們想要利用機器學習來標定臉部特徵點，以輔助兒童睡眠呼吸中止症的判斷。本論文利用FAN以及AAM的兩個現有人臉辨識之方法予以整合且客製化產生模型，FAN為目前準確度十分高的模型，但因其局部特徵的網路結構設計易使模型陷入局部最優解。AAM為較舊的模型，但因其可使用全局訊息且有人工標註的功能，故我們將兩者結合並提出一個可自動標示臉部特徵點的系統架構。實驗結果顯示，我們的自動化系統，與人工判定的方式，平均誤差大約為 13%。未來我們將持續改善此系統，以協助患者可以在家先行初步判斷。

關鍵字

臉部辨識；人工智慧；機器學習

並列摘要

The analysis of facial features can help the diagnosis of children’s obstructive sleep apnea. Traditionally, facial features are identified manually. In this thesis, we want to use the machine learning technique to identify facial features to assist the diagnosis of children’s obstructive sleep apnea. We integrate two existing AI face recognition methods, FAN and AAM, and customize a new model for our work. FAN is accurate in a local region, while AAM can help optimize globally. Then, based on this model, we propose a system architecture that can automatically label facial features. The experimental results show that the average error of our automatic system is about 13% compared with the manual method. In the future, we will continue to improve the system to help patients to make initial judgments at home.

並列關鍵字

Face Recognition ； Artificial Intelligence ； Machine Learning

參考文獻

[1] Wan-Yi Hsueh, “Association between craniofacial photogrammetry, dental arch morphology, and cephalometry in children with sleep-disordered breathing”,2020.