A new feature extraction method for AI based classification of heart sounds: dual-frequency cepstral coefficients (DFCCs)

Classification of biomedical sounds using Artificial Intelligence (AI), especially the examination of heart sounds, is of great importance. However, existing feature extraction methods often fall short in performance due to their limited capacity for frequency analysis and potential information loss. This study proposes a novel feature extraction model called Dual Frequency Cepstral Coefficients (DFCC). This model utilizes a dual filtering approach that combines Mel and Gammatone filter structures, along with cube root and logarithmic transformations weighted for energy conversion. The DFCC model offers a balanced representation of a wide range of signals by integrating Mel and Gammatone filters, which emphasize low-energy components and reduce information loss through the combined transformations. Additionally, the Discrete Fourier transform (DFT) preserves both amplitude and phase information, enabling a more comprehensive analysis in the time-frequency domain. The proposed method was tested for five classes on the Heartbeat sounds dataset using K-Nearest Neighbors(KNN), Support Vector Machine(SVM) and Convolutional Neural Network(CNN) classifiers. According to the results, the DFCC method achieved significant success in classifying heartbeat sounds by reaching an accuracy rate of 93%. The DFCC model stands out as an effective feature extraction method in the classification of biomedical sounds. Future studies could focus on enhancing the capabilities to other types of biomedical sounds beyond heartbeats like respiratory sounds or gastrointestinal sounds.