Automatic adventitious respiratory sound analysis: A systematic review.

No prior work had resolved the lack of a standardized approach for evaluating automated lung sound analysis systems. Researchers often struggle to compare different algorithms due to inconsistent validation methods and performance metrics. This gap motivated a comprehensive investigation into existing computational techniques for identifying abnormal respiratory events. Prior research has shown that digital tools could potentially support clinical diagnosis for conditions like asthma or pneumonia. However, the field remains fragmented with diverse methodologies and reporting standards across various studies. That uncertainty drove the need for a systematic synthesis of the current literature to establish a baseline. Understanding these variations is vital for advancing the reliability of computer-assisted auscultation in clinical settings. This review addresses the need to organize disparate findings into a cohesive framework for future development.

The aim of this work is to provide a comprehensive review of existing algorithms for the detection or classification of abnormal lung sounds. This study addresses the lack of a standardized approach for evaluating these computational systems. The authors seek to organize the diverse methodologies currently present in the scientific literature. By summarizing these techniques, the researchers establish a necessary baseline for future investigations in the field. The study investigates how various digital tools assist physicians in diagnosing or monitoring conditions like asthma and pneumonia. It also explores the limitations of current research regarding performance reporting and validation. This effort clarifies the current state of the art for developers and clinicians alike. The review provides a structured overview to guide subsequent advancements in computer-assisted respiratory diagnostics.

Review Approach involved a systematic search of English articles published between 1938 and 2016. The investigators queried the Scopus and IEEExplore databases to identify relevant publications. They expanded the collection by manually screening references listed in the retrieved papers. Inclusion criteria required that studies focus on the detection or classification of abnormal lung sounds. Each selected report needed to provide sufficient information to allow for the approximate repetition of the work. The team extracted specific data points regarding the sound types, sensor locations, and analytical features used. They also documented the instrumentation and data management strategies employed by the original authors. Finally, the researchers converted reported performance metrics into standardized accuracy measures to facilitate a cohesive synthesis of the literature.

Key Findings From the Literature indicate that 77 reports met the inclusion criteria for this systematic review. Among these, 55 studies focused on wheeze, representing 71.43% of the total analyzed works. Crackles were the second most common subject, appearing in 40 studies, or 51.95% of the literature. Other sounds, including stridor and rhonchi, were examined in 11.69% of the reports each. The review identified that researchers utilized microphones, stethoscopes, and accelerometers for data collection. Analysis techniques ranged from basic thresholding to complex machine learning algorithms. The authors observed that performance metrics varied significantly due to the lack of established validation standards. Despite these inconsistencies, recent studies showed high agreement with conventional non-automatic identification methods.

Synthesis and Implications reveal that automated detection systems demonstrate strong agreement with traditional manual identification techniques. The authors suggest that digital sound analysis serves as a viable alternative to overcome inherent limitations of human auscultation. These tools provide a promising pathway for enhancing the long-term monitoring of chronic pulmonary diseases. The researchers note that the current lack of standardized validation protocols prevents direct performance comparisons between existing models. Future efforts should prioritize the development of uniform evaluation metrics to ensure consistency across the field. This review provides a foundational summary that helps clarify the current landscape of computational respiratory diagnostics. The findings indicate that while progress is significant, the community must move toward shared benchmarks for future success. The authors conclude that automated methods are poised to play a meaningful role in modern clinical practice.