Analysis of Gastrointestinal Acoustic Activity Using Deep Neural Networks

Abstract

Automated bowel sound (BS) analysis methods were already well developed by the early 2000s. Accuracy of ~90% had been achieved by several teams using various analytical approaches. Clinical research on BS had revealed their high potential in the non-invasive investigation of irritable bowel syndrome to study gastrointestinal motility and in a surgical setting. This article proposes a novel methodology for the analysis of BS using hybrid convolutional and recursive neural networks. It is one of the first methods of using deep learning to be widely explored. We have developed an experimental pipeline and evaluated our results with a new dataset collected using a device with a dedicated contact microphone. Data have been collected at night-time, which is the most interesting period from a neurogastroenterological point of view. Previous works had ignored this period and instead kept brief records only during the day. Our algorithm can detect bowel sounds with an accuracy >93%. Moreover, we have achieved a very high specificity (>97%), crucial in diagnosis. The results have been checked with a medical professional, and they successfully support clinical diagnosis. We have developed a client-server system allowing medical practitioners to upload the recordings from their patients and have them analyzed online. This system is available online. Although BS research is technologically mature, it still lacks a uniform methodology, an international forum for discussion, and an open platform for data exchange, and therefore it is not commonly used. Our server could provide a starting point for establishing a common framework in BS research.

Keywords: bowel sounds; deep learning; gastroenterology; machine learning; neural network; software system; sound analysis; spectrogram.

Figure 1

Flowchart of the method developed. The first optional step of the algorithm is converting audio into a single channel WAV format with 24 bit, 44.1 kHz sampling and represent it as a collection of 2-s recordings. Next, we remove the sounds outside interesting spectra using Fast Fourier Transform and Low Pass Filter. The signal is then converted into a spectrogram and the spectrogram obtained is divided into a sequence of frames. Finally, every frame (spectrogram) is classified using machine learning algorithms, with the returning binary value representing the frame as either bowel sound or noise. We calculate the output from adjacent frames.

Figure 2

Smoothed learning curves representing the accuracy of the linking model on the validation set, depending on the maximum frequency of the spectrogram max_freq.

Figure 3

Smoothed learning curves representing the accuracy of the linking model on the validation set, depending on the window width fft for 3 different values.

Figure 4

Hann and Hamming window. The x-axis represents a number of sample n in signal.

Figure 5

Smoothed learning curves showing the accuracy of the connecting model on the validation set, depending on the number of frames.

Figure 6

CRNN model, combination of convolutional layers (a feature extractors) and recursive layers.

Figure 7

Classification model combining a convolutional network and dense network, to classify using adjacent frames results, called the CDNN.