An Integrated System for the Assessment of Carotid Plaque Stability Based on the Artificial Intelligence

Stroke is a global disease with high mortality and disability. Evaluation of the stability of carotid plaque has important clinical significance for stroke prevention.

As a practical, fast, non-invasive, non-radiative inspection technique, ultrasound is widely used in the evaluation of carotid plaque stability. There are many methods for evaluating the stability of carotid plaque by ultrasound. The texture features extracted from two-dimensional ultrasound images can be used to quantify subtle differences in follow up scanning. The elastography of the plaques can be used to quantify changes in the elastic properties of reactive plaques. Contrast enhanced ultrasound (CEUS) can be used to quantify the density of microvessels. However, the results obtained by these different methods are biased and show poor agreement. We lack a unified, systematic evaluation system and objective guideline for plaque stability assessment. In this context, the topic of plaque stability assessment has been intensively explored by researchers from all over the world. Based on state-of-the-art assessment, we found that some plaques that were judged stable on ultrasound did eventually rupture. On the contrary, some unstable plaques on ultrasound did not cause stroke. This confusion indicates that plaque rupture is not only related to its intrinsic characteristics, such as size, shape, composition etc., but also to external factors such as biomechanics, strain, flow field, etc.. In order to achieve a better understanding of the pathological plaque behaviour, it is necessary to study the hemodynamic problems in the arteries, such as wall shear stress, flow separation, and secondary flow. The recent rapid developments in artificial-intelligence techniques have been widely used in medical imaging and image analysis. These techniques allow supervised learning taking full advantage of the available big data. Therefore, the main purpose of this study is to use artificial intelligence technique to i) evaluate a number of indicators including plaque histology, hemodynamics, etc., ii) establish a uniform and quantitative criteria for judging the stability of carotid plaque.

The key scientific problem to be solved in this project is to establish an objective model for predicting the risk of carotid plaque rupture for clinical decisionmaking.

It can be mainly divided into three main modules: database establishment, model training and clinical validation.

1. Database establishment
2. Obtaining clinical information of the patients from the electronic medical record system.
3. Use of image analysis techniques in ultrasound images to study the relationship between different image texture parameters and plaque histological composition. A method for evaluating the stability of carotid plaque based on texture parameters will be established.
4. Use of ultrasound (shear-wave) elastography to establish a method for assessing the carotid-plaque vulnerability based on the assessment of tissue elasticity.
5. Combining carotid velocity vector imaging with the blood-flow vortex manipulation technique to achieve an accurate measurement of the central position of the vulnerable plaque blood flow vortex and the orbital angular momentum, providing a new technique for vulnerable plaque detection based on vortex vector blood flow. Block detection new technology. The biomechanical characteristics of arterial plaques will be evaluated by detecting the synchrony and coordination of the arterial wall. The relationship between the biomechanical characteristics and stability of different types of plaques in the cardiac cycle will also be investigated.
6. Use of quantitative CEUS analysis and localization microscopy to study the relationship between CEUS and neovascularization in carotid plaques, focusing on the influence of microvessel density, architecture, and other quantitative features extracted from CEUS. Based on these features, a prediction model for plaque rupture will be designed and evaluated.
7. Use of computational fluid dynamics modeling to study the effects of geometrical changes on local hemodynamics and blood component concentration distribution during plaque growth. Further analysis of the growth mechanism of plaque and its effect on hemodynamics will be made.
8. Deep-learning model After obtaining all the parameters mentioned above, the correlation between these parameters and carotid plaque stability will be studied through machine learning.

A machine-learning classifier will be designed that takes the extracted features as input. Feature selection will be implemented through k-fold cross-validation to identify the set of complementary features that most contribute to assess plaque stability and predict plaque rupture. In particular, we aim at predicting whether a stroke will occur within 1 year. When sufficient data is available, a deep convolutional neural network will also be designed and trained to learn the prediction model. In this process, the risk of overfitting will be carefully considered by evaluating the classification performance over training, test, and validation sets.

As a result, the model complexity will be adjusted to the available training sets.

1. clinical evaluation The validation of the developed stroke prediction model from rupture of carotid plaques will be extended to a number of hospitals. The model reliability and usability will be tested. Model update and improvement is envisaged in a cyclic fashion, profiting from clinical results and feedback. The accuracy rate auxiliary decision support is expected to levels that are not lower than 90%.