Practical implementation of artificial intelligence algorithms in pulmonary auscultation examination

Area of Science:

• Medical Technology
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Lung auscultation is subjective and highly dependent on physician experience.
• Existing AI solutions for auscultation use limited datasets and unreliable validation methods.

Purpose of the Study:

• To develop and evaluate a novel AI system for objective analysis of respiratory sounds.
• To compare the diagnostic performance of the AI system against human physicians in identifying lung auscultation sounds.

Main Methods:

• An electronic stethoscope captured respiratory sounds from pediatric patients.
• A neural network (NN) based algorithm was trained on thousands of real patient signals.
• A blind test compared AI algorithm performance with a group of physicians on 522 auscultatory sounds.

Main Results:

• The AI algorithm achieved high recall (sensitivity) and F1-score in detecting wheezes, rhonchi, and fine/coarse crackles.
• Machine learning-based analysis proved more efficient than human interpretation for all four sound classes.
• AI demonstrated statistically proven high performance in identifying pathological respiratory sounds.

Conclusions:

• AI-driven automatic sound analysis significantly enhances the efficiency of lung auscultation.
• Implementing AI can minimize errors in interpreting auscultation sounds.
• This technology offers a reliable and objective method for respiratory sound analysis.