(A-2) Uterine contraction detection using multi-scale sliding window Gaussian mixed model.

Introduction:: Electromyometrial imaging (EMMI) is an effective method to non-invasively monitor uterine contraction. It collects the electrical signal from the patient’s body surface and then reconstructs the electrical activity on the uterine surface through inverse computation. However, uterine contraction is challenging to detect due to low Signal-to-noise ratio. Therefore, we need an effective tool to segment the electrical signal on the uterine surface into contraction and non-contraction parts. To address the challenge, we proposed a muti-scale sliding window Gaussian mixed model to combine the information from multiple scales to effectively detect uterine contractions.

Materials and Methods:: Electrical signal on the uterine surface was segmented into many small parts using overlapped sliding window approach. The energy for each segment was then computed. It was hypothesized that the energy of contraction and non-contraction parts followed two different Gaussian models. The expectation-maximization (EM) algorithm was used to fit the Gaussian mixed model. To combine the information from different scales, sliding windows with different sizes were used to segment the signal respectively. A binary feature map can be generated corresponding to each sliding window of a certain scale. Feature maps from different scales could be stacked into a new feature map which could effectively capture the essence to differentiate real and fake contraction. Morphological operations were applied to remove the features of fake contractions, and the location of real contractions could be detected on the resulting feature map.

Results, Conclusions, and Discussions:: Results and discussion: The result of contraction detection on the uterine surface was shown in Figure 1c). The result is compared to Toco monitor to validate that false alarm is removed by combining multi-scale information from the binary feature map.

Conclusions: Multi-scale sliding window Gaussian mixed model can avoid the false alarm by combining information from multi-scale and effectively detect the contraction on the uterine surface.


Acknowledgements (Optional): :

References (Optional): :