Feasibility of AI-based Classification of Normal, Wheeze and Crackle Sounds From Stethoscope in Clinical Settings

Not Applicable

Completed

• Conditions: Respiratory; Lung

• Registration Number: NCT05268263
• Lead Sponsor: Innova Smart Technologies (Pvt.) Ltd

Brief Summary

Assessing the feasibility and testing the accuracy of the developed artificial intelligence algorithms for detection of wheezes and crackles in patients with lung pathologies in clinical settings on unseen local patient data acquired through three digital stethoscopes.

Detailed Description

Not available

Study Timeline

• Study Start: 2022-01-06
• Study First Submitted: 2022-01-08
• Primary Completion: 2022-02-22
• Study Completion: 2022-02-22
• Study First Submitted QC: 2022-02-22
• Study First Posted: 2022-03-07
• Status Verified: 2023-04
• Last Update Submitted: 2023-04-04
• Last Update Posted: 2023-04-06

Recruitment & Eligibility

• Status: COMPLETED
• Sex: All
• Target Recruitment: 60

Inclusion Criteria

• Ages all
• Written consent provided

Exclusion Criteria

• Subject condition unstable
• Chest wall deformity or wounds in adhesive application areas
• Written consent not provided

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: SINGLE_GROUP

Primary Outcome Measures

Name	Time	Method
Clinical validation of AI models for detection of wheeze, crackles, and normal lung sounds by comparison with gold standard	2 months	AI models will be tested for their clinical feasibility through comparison of results obtained from AI models with that of the gold standard by measuring positive and negative agreement (NPA \& PPA). The gold standard is the label given to each lung sound recording by an experienced consultant pulmonologist. The AI model is blinded to these labels and is tested independently for detection of normal lung sounds, wheezes, and crackles
Testing the accuracy of artificial intelligence models for detection of wheeze, crackles, and normal lung sounds by measuring the sensitivity and specificity	2 months	Artificial intelligence models are trained on lung sounds collected from three different digital stethoscopes named NoaScope, eSteth, and Littmann individually. Data from all three digital stethoscopes is also merged to train separate AI models. These trained AI models will be evaluated based on sensitivity which is the ability to correctly identify wheezes and crackles, and specificity which is the ability to correctly identify normal lung sounds. True positive (TP), true negative (TN), false positive (FP), and false-negative (FN) values will be used to calculate sensitivity \& specificity using the following expressions.; Sensitivity: TP/TP+FN Specificity: TN/TN+FP

Secondary Outcome Measures

Name	Time	Method
Performance analysis of three digital stethoscopes: Littmann, NoaScope, and eSteth	2 months	Performance analysis of three digital stethoscopes NoaScope, eSteth, and Littmann will be evaluated using the sensitivity and specificity achieved by each stethoscope. True positive (TP), true negative (TN), false positive (FP), and false-negative (FN) values will be used to calculate sensitivity \& specificity using the following expressions.; Sensitivity: TP/TP+FN Specificity: TN/TN+FP

Trial Locations

Locations (1)

Lady Reading Hospital, Pakistan; 🇵🇰 Peshawar, Pakistan