Computer aided diagnosis of cerebrovascular disease based on DSA image
Shi, Keke (2021-07-12)
Computer aided diagnosis of cerebrovascular disease based on DSA image
(12.07.2021)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2021081142870
https://urn.fi/URN:NBN:fi-fe2021081142870
Tiivistelmä
In recent years, the incidence of cerebrovascular diseases in China has shown a significant upward trend, and it has become a common disease threatening people's lives. Digital Subtraction Angiography (DSA) is the gold standard for the diagnosis of clinical cerebrovascular disease, and it is the most direct method to check the brain lesion. At present, there are the following two problems in the clinical research of DSA images: DSA is a real-time image with numerous frames, containing much useless information in frames; thus, human interpretation and annotation are time-consuming and labor-intensive. The blood vessel structure in DSA images is so complicated that high practical skills are required for clinicians. In the computer-aided diagnosis of DSA sequence images, there is currently a lack of automatic and effective computer-aided diagnosis algorithms for cerebrovascular diseases. Based on the above issues, the main work of this paper is as follows:
1.A multi-target detection algorithm based on Faster-RCNN is designed and applied to the analysis of brain DSA images. The algorithm divides DSA images into arterial phase, capillary phase, pre-venous phase and sinus phase by identifying the main blood vessel structure in each frame. And on this basis, we analyze the time relationship between the time phases.
2.On the basis of DSA phase detection, a key frame location algorithm based on single blood vessel structure detection is designed for moyamoya disease. First, the target detection model is applied to locate the internal carotid artery and the Willis circle. Then, five frames of images are extracted from the arterial period as keyframes. Finally, the nidus' ROI is determined according to the position of the internal carotid artery.
3.A diagnostic method for cerebral arteriovenous malformation (AVM) is designed, which combines temporal features and radiomics features. First, on the basis of DSA time phase detection, we propose a deep learning network to extract vascular time features from the DSA video; then, the time feature is combined with the radiomics features of the static keyframe to establish an AVM diagnosis model. While assisting diagnosis, this method does not require any human intervention, and reduces the workload of clinicians. The diagnostic model that combines time features and radiomics features is applied to the study of AVM staging. The experimental results prove that the classification model trained by fusion features has better diagnostic performance than the model trained by either time features or radiomics features.
Based on the above three parts, this paper establishes a cerebrovascular disease analysis framework based on radiomics method and deep learning. We introduce corresponding solutions for DSA automatic image reading, rapid diagnosis of moyamoya disease, and precise diagnosis of AVM. The method proposed in this paper has practical significance for assisting the diagnosis of cerebrovascular disease and reducing the burden of medical staff. Digital Subtraction Angiography(DSA), Radiomics analysis, Arteriovenous malformations, Moyamoya, Faster-RCNN, Temporal features, Fusion features
1.A multi-target detection algorithm based on Faster-RCNN is designed and applied to the analysis of brain DSA images. The algorithm divides DSA images into arterial phase, capillary phase, pre-venous phase and sinus phase by identifying the main blood vessel structure in each frame. And on this basis, we analyze the time relationship between the time phases.
2.On the basis of DSA phase detection, a key frame location algorithm based on single blood vessel structure detection is designed for moyamoya disease. First, the target detection model is applied to locate the internal carotid artery and the Willis circle. Then, five frames of images are extracted from the arterial period as keyframes. Finally, the nidus' ROI is determined according to the position of the internal carotid artery.
3.A diagnostic method for cerebral arteriovenous malformation (AVM) is designed, which combines temporal features and radiomics features. First, on the basis of DSA time phase detection, we propose a deep learning network to extract vascular time features from the DSA video; then, the time feature is combined with the radiomics features of the static keyframe to establish an AVM diagnosis model. While assisting diagnosis, this method does not require any human intervention, and reduces the workload of clinicians. The diagnostic model that combines time features and radiomics features is applied to the study of AVM staging. The experimental results prove that the classification model trained by fusion features has better diagnostic performance than the model trained by either time features or radiomics features.
Based on the above three parts, this paper establishes a cerebrovascular disease analysis framework based on radiomics method and deep learning. We introduce corresponding solutions for DSA automatic image reading, rapid diagnosis of moyamoya disease, and precise diagnosis of AVM. The method proposed in this paper has practical significance for assisting the diagnosis of cerebrovascular disease and reducing the burden of medical staff.