Automatic detection approach for bioresorbable vascular scaffolds using a u-shaped convolutional neural network
Artificial stent implantation is one of the most effective ways to treat vascular diseases. However, commonly used metal stents have many negative effects, such as being difficult to remove and recover, whereas bio-absorbable stents have become the best way to treat vascular diseases because of thei...
Saved in:
Main Authors: | , , , , , , , |
---|---|
Other Authors: | |
Format: | Article |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/103299 http://hdl.handle.net/10220/49966 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Artificial stent implantation is one of the most effective ways to treat vascular diseases. However, commonly used metal stents have many negative effects, such as being difficult to remove and recover, whereas bio-absorbable stents have become the best way to treat vascular diseases because of their absorbability and harmlessness. It is very important in vascular medical imaging, such as optical coherence tomography (OCT), to be able to effectively track the position of stents in blood vessels. This task is undoubtedly labor-intensive, and it is inefficient to rely on experts to identify various scaffolds from medical images. In this paper, a novel automatic detection method for bioresorbable vascular scaffolds (BVSs) via a U-shaped convolutional neural network is developed. The method is composed of three steps: data preparation, network training, and network testing. First, in the data preparation step, we complete the task of labeling related samples based on expert experience, and then, these labeled OCT images are divided into the original and masked OCT images (corresponding to X and Y in supervised learning, respectively). Next, we train our data on a U-shaped convolutional neural network, which consists of five downsampling modules and four upsampling modules. We can obtain a related training model, which can be used to predict the related samples. In the testing stage, we can easily utilize the trained model to predict the input OCT data so that we can obtain the relevant information about a BVS in an OCT image. Obviously, this method can assist doctors in diagnosing the disease and in making important decisions. Finally, some experiments are performed to validate our proposed method, and the IoU criterion is used to measure the superiority of our proposed method. The results show that our proposed method is completely feasible and superior. |
---|