Fig 1.
Schematic diagram summarizing the course of the study.
It starts by recording the BS by placing the sensor on the subjects’ abdomen, proceeds to labeling the signal into non-BS and BS patterns (SB, MB, CRS, HS), and segmenting the signal into 2 seconds overlapped windows, to use it later on the classification with 3 different methods (using tabular features, using spectrogram and using pre-trained models).
Fig 2.
Bowel sound patterns examples extracted from the dataset used in this study, the left column represents the signal in the time domain, and the right column describes the signal in the frequency domain.
Starting from the top, (a) Single Burst (SB), (b) Multiple Burst, (c) Continuous Random Sound (CRS), and (d) Harmonic Sound (HS).
Fig 3.
Distribution of bowel sound pattern (SB, MB, CRS, HS) counts by subjects, the box represents the interquartile range (IQR), with the horizontal line inside the box indicating the median.
Whiskers extend to 1.5 IQR, and points outside the whiskers represent outliers. The SB group shows the highest median and variability, while the HS group has the smallest counts and minimal variability.
Fig 4.
The first column shows the distribution of the classes (non-BS, SB, MB, HS) within the dataset before and after the segmentation using 2 2-second, overlapping window and results in a reduced count of patterns, particularly for short-duration patterns like SB and HS that are shorter than the window size.
The second column shows the distribution of the 5 classes within the train, validation, and test sets by using random and stratified splits. The distribution of the classes is closer to the required ratio (70%, 15%, 15%) using stratified splitting.
Fig 5.
The AUC values of the binary classification between non-BS and BS signal are listed as follows: using tree-based models on tabular features(green; bars from 1 to 4), Spectrogram-based models (blue; bars from 5 to 16), and transfer learning based features (orange; bars from 17 to 19).
The figure demonstrates that pretrained models offer more reliable performance compared to other techniques, while feature-based models struggle with the bowel sound detection task.
Fig 6.
The binary AUC values for each class vs the rest (NONE vs all, SB vs all, MB vs all, CRS vs all, and HS vs all) and the overall AUC, achieved by the best models within each technique group.
This diagram highlights that pretrained models provide stable performance across all bowel sound pattern classes, whereas other techniques face greater challenges, particularly with patterns that have fewer samples, such as SB and HS.
Table 1.
The binary AUC values for each class vs the rest (NONE vs all, SB vs all, MB vs all, CRS vs all, and HS vs all) and the overall AUC, using tabular features, spectrograms, and transfer learning-based models.
The best performance achieved by the Wav2Vec model for all classes.