SCH: Smart Auscultation for Pulmonary Diagnostics and Imaging

SCH：用于肺部诊断和成像的智能听诊

• 批准号: 10590732
• 负责人: Mounya Elhilali
• 金额: $ 29.28万
• 依托单位: JOHNS HOPKINS UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-03-15 至 2026-02-28
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10590732
• 关键词: Accident and Emergency department; Acoustics; Address; Affect; Air; Air Movements; Alveolus; Ambulances; Anatomy; Architecture; Artificial Intelligence; Auscultation; Biological Markers; Blood Tests; Breathing; Case Management; Case/Control Studies; Categories; Chest; Childhood; Clinic; Clinical; Clinical Data; Communities; Complement; Complex; Consumption; Devices; Diagnosis; Diagnostic; Ear; Elements; Engineering; Environment; Goals; Hand; Head; Health; Health Personnel; Heart; Home; Hospitals; Image; Imaging Device; Inflammation; Interobserver Variability; Judgment; Location; Lower Respiratory Tract Infection; Lung; Lung infections; Machine Learning; Masks; Medical; Modality; Monitor; Mucous body substance; Noise; Nursing Staff; Obstruction; Outcome; Oxygen saturation measurement; Patients; Pattern; Physicians; Pneumonia; Positioning Attribute; Public Health; Radiation; Resources; Respiratory Diaphragm; Respiratory Sounds; Respiratory System; Respiratory Tract Infections; Roentgen Rays; Scheme; Signal Transduction; Skin; Source; Standardization; Stethoscopes; Techniques; Technology; Telemedicine; Testing; Thoracic Radiography; Time; Training; Travel; Triage; Visualization; accurate diagnosis; clinical biomarkers; clinical diagnosis; clinical examination; clinical practice; community clinic; cost; deep learning; design; diagnostic value; electric impedance; field study; frontier; imaging modality; improved; innovation; interest; invention; lung imaging; medical attention; new technology; novel; pediatric emergency; point of care; recurrent neural network; respiratory health; screening; sensor; sound; standard care; tool; transmission process; treatment strategy; ultrasound; usability; virtual; wearable device

The stethoscope is a ubiquitous technology used to listen to sounds from the chest in order to assess lung or heart conditions. Despite its universal use, it is considered an unreliable diagnosis tool due to a number of limitations: masking by noise, need for highly trained users and ear to interpret lung sounds and subjectivity in interpreting auscultation sounds. Still, one of the reasons auscultations are a staple of clinical screening is that sound is one the cheapest, fastest and most readily available biomarkers. The simple fact of breathing involves sound traveling through chest cavities that will be affected by presence of obstructions or abnormalities. While the signature of these air flow disruptions may be concealed, the right engineering innovation should not only identify their presence but can be extended as an imaging modality to identify their location, which would be a novel use of breath sounds to image lung cavities. The proposed smart auscultation technology is innovative in three ways: (i) it develops a machine learning architecture that imposes finite-element airway propagation constraints and stochastic variational inference using recurrent neural networks, (ii) a novel piezo-sensing material with tunable acoustic impedance that matches the skin hence eliminating air as transmission medium between the chest and device diaphragm which virtually eliminates pick up of any ambient noise, (iii) an array device that leverages the piezo-sensor to develop an imaging device using passive breathing sounds (instead of radiations or ultrasound probes). The proposed technology is extremely low-cost, deployable under adverse conditions, usable for immediate clinical examination as well as extendable for monitoring as a wearable device. The new technology will be field tested directly in case/control studies at the Johns Hopkins pediatric ER and pulmonary clinics to validate localization accuracy from the auscultation array using physicians’ judgments as gold standard. If successful, this technology will complement alternative, often costly and time-consuming diagnosis schemes (X-rays or ultrasounds which often cost $100-$1000’s) to offer a fast, cheap (few $) and accessible tool that can be widely disseminated from community clinics to hospitals and potentially home-based health monitoring. Given the dire public health need in addressing ALRI challenges, the proposed low-cost and efficient technology can be a game changer as a point-of-care aid to triage cases that require further medical attention.

听诊器是一种无处不在的技术，用于聆听胸部的声音，以评估肺部或心脏状况。尽管它被广泛使用，但由于许多限制，它被认为是一种不可靠的诊断工具：被噪音掩盖，需要训练有素的用户和耳朵来解释肺部声音以及解释听诊声音的主观性。尽管如此，听诊仍然是临床筛查的主要方法之一，因为声音是最便宜、最快、最容易获得的生物标志物之一。呼吸的简单事实涉及声音通过胸腔传播，这将受到阻塞或异常的影响。虽然这些气流中断的特征可能会被隐藏，但正确的工程创新不仅应该识别它们的存在，而且可以扩展为成像模式以识别它们的位置，这将是呼吸声成像肺腔的新用途。所提出的智能听诊技术在三个方面具有创新性：（i）它开发了一种机器学习架构，该架构使用递归神经网络来施加有限元气道传播约束和随机变分推理，（ii）一种新颖的压电-具有匹配皮肤的可调声阻抗的传感材料，因此消除了空气作为胸部和设备隔膜之间的传输介质，这实际上消除了任何环境噪声，（iii）利用压电传感器来开发使用被动呼吸声（而不是辐射或超声探头）的成像设备的阵列设备。所提出的技术成本极低，可在恶劣条件下部署，可用于立即临床检查，并可作为可穿戴设备扩展用于监测。这项新技术将在约翰霍普金斯儿科急诊室和肺科诊所的病例/对照研究中直接进行现场测试，以医生的判断作为金标准，验证听诊阵列的定位准确性。 如果成功，这项技术将补充替代的，往往是昂贵和耗时的诊断方案（X光或超声波，往往花费100美元至1000美元），以提供一个快速，廉价（几美元）和方便的工具，可以广泛传播从社区诊所到医院和潜在的家庭健康监测。鉴于在应对ALRI挑战方面的迫切公共卫生需求，拟议的低成本和高效技术可以改变游戏规则，作为对需要进一步医疗关注的病例进行分类的护理点援助。