Leveraging Deep Learning Models for Classification of Inhalation and Exhalation in Respiratory Cycles Using Thermal Images

Breathing Pattern Disorders (BPD) significantly impact the quality of life for many individuals, particularly those suffering from asthma and anxiety-related conditions. These disorders can lead to chronic respiratory issues, reduced physical performance, and heightened psychological stress. Therefore, accurately detecting and monitoring breathing patterns is crucial for effectively managing and treating BPD. To address this issue, this paper proposes several deep learning models, including DenseNet121, VGG16, ResNet152V2, InceptionV3, and NAS-NetMobile, to classify inhalation and exhalation phases within respiratory cycles using thermal imaging. By leveraging transfer learning, distinct breathing patterns, including long, normal, and rapid cycles, were effectively identified from thermal signatures. We have evaluated the performance of the proposed deep learning models using a labeled thermal image dataset. The results show the effectiveness of the proposed models, specifically VGG16 and DenseNet121, which demonstrated high classification accuracies of 96% and 93%, respectively. This non-invasive approach enhances diagnostic capabilities in detecting and monitoring BPD. The integration of automated image classification into respiratory analysis offers a promising tool for healthcare professionals to assess respiratory health and potentially preempt clinical deterioration through early intervention.