Tri-MTL: A Triple Multitask Learning Approach for Respiratory Disease Diagnosis

Area of Science:

• Medical informatics
• Artificial intelligence in healthcare
• Respiratory medicine

Background:

• Auscultation is vital for diagnosing respiratory conditions, relying on lung sounds, patient history, and test results.
• Multitask learning (MTL) shows promise in integrating diverse medical data but has not fully explored the interplay of respiratory sounds, disease manifestations, and patient metadata.
• A research gap exists in leveraging deep learning and MTL to simultaneously analyze respiratory sounds and associated clinical data.

Purpose of the Study:

• To investigate the integration of MTL with deep learning architectures for enhanced respiratory sound classification and disease diagnosis.
• To evaluate the impact of incorporating patient metadata (e.g., stethoscope information) within an MTL framework.
• To improve the accuracy and efficiency of diagnostic decision-making in respiratory care.

Main Methods:

• Utilized multitask learning (MTL) combined with advanced deep learning models.
• Integrated patient metadata, including stethoscope information, into the MTL framework.
• Conducted comprehensive experiments to assess performance in lung sound classification and disease diagnosis.

Main Results:

• Significant improvements observed in both lung sound classification and disease diagnostic performance.
• Incorporating stethoscope information into the MTL architecture led to enhanced results.
• Achieved high specificity (85.83% for diagnosis, 78.86% for classification) and sensitivity (94.09% for diagnosis, 41.56% for classification).

Conclusions:

• The integrated MTL approach effectively enhances respiratory sound classification and disease diagnosis.
• Patient metadata plays a crucial role in improving diagnostic accuracy within the MTL framework.
• This approach offers immediate clinical applications to support medical professionals, potentially reducing misdiagnosis and improving patient outcomes in respiratory care.