Objective:

To develop a deep learning method based on high-resolution network (HRNet) for key point detection of the tip depth of infusion port catheter in radiograph. The proposed method involves the use of HRNet to learn the features of the images and accurately detect the tip position of the catheter. The automated system will also provide the optimal placement distance for the cathete

r and discussing its potential applications.

Methods:

A total of 530 chest X-ray images with the catheter tip of the infusion port were selected

and the catheter tip

tracheal carina

and the first to second thoracic spine under the tracheal carina were annotated and the model was trained with an improved HRNet. The position of the catheter tip was classified by the key point coordinates of the catheter tip predicted by the model and the upper and lower boundaries of the optimal catheterization area (the upper and lower vertebral spaces of the second vertebral body)

and the vertical distance between the catheter tip and the upper and lower boundaries was calculated. Compared the error of the annotated data with the predicted data and make statistics. The accuracy

precision

specificity

sensitivity and F1 score of classification were analyzed by confusion matrix.

Results:

The average error between the key points predicted by the model and the ground truth key points was (3.882±6.568) mm. The distance errors from the tip of the catheter to the upper and lower boundaries of the optimal catheter placement area were (6.584±8.652) mm and (4.249±6.079) mm. The accuracy of model classification was 0.925 (95% CI 0.8901~0.9487)

accuracy was 0.887 (95% CI 0.813~0.934)

sensitivity was 0.887 (95% CI 0.813~0.934)

specificity was 0.943 (95% CI 0.904~0.967)

and F1 score was 0.887.

Conclusion:

It is feasible to use the key point detection technology based on HRNet to detect the tip of infusion port catheter andclassify and measure the position of the catheter. This study showed that the key point detection technology can improve the quality of catheters during the installation of infusion port.