Research On The Segmentation Of Guide-Wire And Catheter Tip Based On X-Ray Fluoroscopy Images

The morbidity and mortality of coronary heart disease are gradually rising with the development of society.The age group of people suffering from coronary stenosis is gradually becoming younger,which is a major public health problem in society.Percutaneous coronary intervention is one of the main methods for the treatment of coronary heart disease.During the interventional treatment,doctor observes the current shape and position of the interventional medical devices such as catheter and guide-wire in the X-ray fluoroscopy images generated by the C-arm X-ray machine in real-time.It is of great significance to clearly display the position and shape of interventional devices such as guide-wire and catheter in fluoroscopy images.In addition,by tracking the position of the catheter tip placed at the coronary ostium in the X-ray fluoroscopy image sequence,the image displacement information caused by the heart beating can be obtained,which can help the dynamic coronary roadmapping to be overlaid on the X-ray fluoroscopy images.Therefore,the current academia and industry have great research prospects and application requirements for low false detection rate,low missed detection rate,and high robustness of guide-wire segmentation algorithms and catheter tip segmentation algorithms under X-ray fluoroscopy images.There are differences between the characteristics of the guide-wire and the catheter tip in X-ray fluoroscopy images,so two different network structures are designed respectively for the guide-wire segmentation task and the catheter tip segmentation task.According to the guidewire in the X-ray fluoroscopy image has the characteristics of slenderness,a deep-learning based guide-wire segmentation network with Transformer’s global self-attention is designed.The guide-wire segmentation network’s Dice coefficient is improved by an average of 3.2%,the accuracy rate is improved by an average of 2.1%,and the recall rate is improved by an average of 4.6% on the clinical X-ray fluoroscopy image dataset.According to the catheter tip has the characteristics of simple and small,a light-weight catheter tip segmentation network is designed,and the self-distillation method is used to make up for the decline in accuracy caused by the light-weight model.The light-weight catheter tip segmentation network’s accuracy rate is the same as that of large networks,the inference speed is increased by 8.72 ms,and the number of model parameters is reduced by 1.2 M.Due to the problems of poor contrast and unclear display of the guide-wire and the catheter tip in X-ray fluoroscopy images,this paper also designs corresponding post-processing methods for these two segmentation tasks.According to the linear feature of the guide-wire,the bidirectional Dijkstra shortest path algorithm is introduced,and an automatic endpoint location algorithm is designed to realize the end-to-end guide-wire segmentation post-processing algorithm.Aiming at the problem of the high false detection rate of catheter tip caused by contrast agent residue,the Bayesian filtering algorithm is improved to make full use of timing information and deal with the existence of multiple catheter tips.The improved Bayesian filtering algorithm reduces the image false detection rate by 2.36% and the sequence false detection rate by 2.08%.