Machine Learning and Cardiology

Fetal Peak Detection - Results & Conclusions

Detection was done in a qualitative style. Once the peaks were discovered, the eye was the discerning factor in testing the accuracy of detection. Fetal peaks were detected quite easily except when the analysed fetus was very young (22-23 weeks old). Lower accuracy occurred when peaks were obtained which authenticity was doubtful, such as when they occurred on the same time as the mother's heartbeat or when four beats were found within the space of two. When the peaks were too erratic to be anything else but noise, no results were found at all. Table 5.1, tabulates the accuracy indicators given after the test data was run through fetal peak detection, very good meaning regular heartbeats could be found.

fig 5.1: Peak Detection Results

 

5.1 Discussion

Judgement about qualitative analysis of the results is probably the issue which could be most argued about in this research. Usually it is the case that the best results are those that can be put to mathematical tests, statistics and error analyses. Here, results are given in the form of unclear, okay, good and very good. However, often research results in this field are open to interpretation. In such cases, experts are the deciding factor in the outcomes of the used techniques. Therefore, a better understanding in the dataset would have helped in a better representation of the results. For example, there are no annotations or descriptions of problems with the fetus whether it is healthy. This was an important barrier to overcome for our test data, as it was assumed that all fetuses were considered having normal heartbeats.

Taking this knowledge into account, the used techniques provided promising results. The research showed that a fetal heartbeat could be found even in early stages of development and that it was able to pick up most peaks only missing one or two (in a series of around 23 fetal heartbeats). Moreover, the location of the missing peaks is often very clear, and some more advanced techniques would probably allow to also reconstruct them. The peaks that are missing are mainly caused by either too much noise or by simultaneous appearance with the maternal peaks, giving doubt as to whether fetal peaks truely exist there or not.

 

5.2 Conclusions

As stated above, the peak extraction process is very clear in its results. This shows that the methods followed give a good indication on the whereabouts of fetal peaks in these sorts of ECGs. Improvements are needed in the preprocessing of the data mainly in reducing the noise, although this is difficult due to the weakness of fetal signals. From here it is mainly getting expert opinion on the results on not layman of this medical field. Although the analysis was indicative of good results, in the end only an expert could give any definitive answer.

 

5.3 Bibliography

 

<<< Previous | Index | Next >>>