Published Paper Details:

Recognition of respiratory conditions at the right time remains fundamental to deliver both proper treatment and superior patient results. Standard auscultation techniques face two significant problems due to subjective evaluation of patients' airway sounds and inconsistent interpreter interpretations. The system presents a real-time intelligent phonopulmography system with acoustic sensing capabilities to automate the precise classification of lung sounds. The system uses special hardware components built from a combination of stethoscope along with microphone and operational amplifier circuits connected to an Arduino to detect and improve respiratory audio signals. Through MATLAB processing the acquired sounds become structured data that can be classified using methods such as Mel-frequency cepstral coefficients and spectral analysis during feature extraction. A deep learning-based version of the convolutional neural network runs in Python using TensorFlow to classify lung sounds into the categories of normal along with crackles, wheezes and rhonchi. A pre-trained feature extractor enables performance improvement in the system. The model processed 3,680 clinically marked audio samples to reach more than 90% successful classifications. Modern usability features of the web interface allow real-time audio recording with automatic diagnostic assessment following AI model processing. The research demonstrates how coupled sensing hardware with AI methods accomplishes real-time respiratory evaluation that is both reliable and scalable. Upcoming developmental goals target the expansion of the database as well as the improvement of response speeds for future mobile and cloud-based remote healthcare applications.