心音识别的机遇与挑战:深圳心音数据库简介

Opportunities and Challenges for Heart Sound Recognition:A Brief on the Heart Sounds Shenzhen Corpus

随着可穿戴设备与人工智能技术的发展,开发1种基于机器听觉理论的“非侵入式”心音识别系统显得尤为重要.然而,当前研究的局限在于可用于相关研究的公开的标准化数据库资源稀缺.并且,相关的研究成果存在可复现性不高、未考虑受试者独立性、测量指标不统一等缺陷.本文旨在介绍深圳大学总医院录制的全新标准化心音识别数据库——Heart Sounds Shenzhen(HSS),并讨论官方给出的标准化方法和基准线,指出心音识别领域的机遇与挑战,以期得到国内外同行的关注与后续的深入研究.

With the development of wearables and artificial intelligence,developing a non-invasive heart sound recognition system via the machine listening theory becomes a vital research topic.Nevertheless,a standard publicly accessible heart sound corpus is urgently lacking in the current studies.In addition,low reproducibility,subject-dependency,and inconsistent evaluation metrics prevail in the existing reported results.To this end,we introduce a most recent standard heart sound corpus,the Heart Sounds Shenzhen(HSS)database.We discuss the state-of-the-art methods and the baseline as featured in an official competition,and indicate the opportunities and challenges.We hope this contribution can attract more attention and further studies to this relevant area.