Method Cazares et al. (2002)
Cazares, Shelley Marie - Automated identification of abnormal patterns in the intrapartum cardiotocogram Ph.D. Thesis, University of Oxford , Oxford,2002
This theses has been realised at the Oxford University on the team of Dawes and Redman. Dawes and Redman had already developped a method to realise morphological analysis [Dawes et al. 1982] of the Fetal Heart Rate and they have improved it on [Pardey et al. 2002]. In parrallel of Pardey, the thesis of S. Cazares proposed an other method to realise this morphological analysis. It seems the method commercially distributed on the software Sonicaid Fetalcare is based on the method of Cazares but there is numerous unpublished improvement.
The Cazares method consists in using morphological filter to remove accident on the FHR and thus get the baseline. Morphological filter resides on two operators : erode and dilate. An erosion consists in taking the minimimum of the signal on a sliding window :
Whereas a dilatation consists in taking the maximum of the signal on a sliding window :
Then we can define an opening which is the result of an erosion followed by a dilation () and a closing which is the result of a dilation followed by an erosion (). The figure  illustrate the results of those operators on FHR signal. I can be seen that an opening filter can be used to remove acceleration and the closing filter can be used to remove deceleration.
The final recommanded method consists in using an opening followed to a closing to remove both acceleration and deceleration, and then a smoothing by a sliding average.
The idea seems pertinent since it is a simple method to remove accident while being able to follow baseline variation. The main limitation however is the prolonged accident which cannot be removed if the duration is longer than the window of opening/closing. On the other side, if the windows is too large the accident can be merged and thus they can be after considered as baseline. A more global analysis (on larger window) of the FHR should be done on complicated case to determine the baseline. Maybe better results could be obtain by realising successive opening/closing and closing/opening with length increasing progressively, and taking average of the results of both opening/closing and closing/opening