Non-invasive Detection of Bowel Sounds in Real-life Settings Using Spectrogram Zeros and Autoencoding

Gastrointestinal (GI) diseases are amongst the most painful and dangerous clinical cases, due to inefficient recognition of symptoms and thus, lack of early-diagnostic tools. The analysis of bowel sounds (BS) has been fundamental for GI diseases, however their long-term recordings require technical and clinical resources along with the patientt’s motionless concurrence throughout the auscultation procedure. In this study, an end-to-end non-invasive solution is proposed to detect BS in real-life settings utilizing a smart-belt apparatus along with advanced signal processing and deep neural network algorithms. Thus, high rate of BS identification and separation from other domestic and urban sounds have been achieved over the realization of an experiment where BS recordings were collected and analyzed out of 10 student volunteers.