Recognition of Heart Murmur Based on Machine Learning and Visual Based Analysis of Phonocardiography

Heart sounds recognition is essential for heart defects diagnosis. Diagnosis of structural heart defects is not always possible using the contemporary stethoscope and need further assessment using high cost devices such as X-ray, electrocardiogram (ECG), echocardiography (ECHO) and computed tomography (CT)). Automatic computer assisted auscultation may be used as a clinical decision support tool. In an attempt to develop an automatic computer aided diagnostic modality for heart conditions that is sensitive, specific, non-invasive we created two automatic computer cardiac auscultation (ACCA) models that provide heart sound analysis and we aimed to improve the sensitivity and correct classification rate (CCR) of recognition of heart sounds, thus we developed model A ACCA recognition system (machine learning (interpreter independent)) and model B ACCA recognition system (machine learning and interpreter dependent visual analysis). We used machine learning based on mel frequency cepestral coefficients as a feature and Hidden Markov Model (HMM) as a classifier. We performed visual analysis based on phonocardiography (PCG) and spectrogram image. Model A ACCA demonstrated 97% CCR, 99.2% sensitivity and specificity 100% and model B ACCA demonstrated 98.2% CCR and 99.2% sensitivity and specificity 100%. ACCA models A and B allow reliable interpretation of recognised sounds to detect underlying structural defects.