An automatic multi-class coronary atherosclerosis plaque detection and classification framework

Abstract

Detection of different classes of atherosclerotic plaques is important for early intervention of coronary artery diseases. However, previous methods focused either on the detection of a specific class of coronary plaques or on the distinction between plaques and normal arteries, neglecting the classification of different classes of plaques. Therefore, we proposed an automatic multi-class coronary atherosclerosis plaque detection and classification framework. Firstly, we retrieved the transverse cross sections along centerlines from the computed tomography angiography. Secondly, we extracted the region of interests based on coarse segmentation. Thirdly, we extracted a random radius symmetry (RRS) feature vector, which incorporates multiple descriptions into a random strategy and greatly augments the training data. Finally, we fed the RRS feature vector into the multi-class coronary plaque classifier. In experiments, we compared our proposed framework with other methods on the cross sections of Rotterdam Coronary Datasets, including 729 non-calcified plaques, 511 calcified plaques, and 546 mixed plaques. Our RRS with support vector machine outperforms the intensity feature vector and the random forest classifier, with the average precision of 92.6 ± 1.9% and average recall of 94.3 ± 2.1%. The proposed framework provides a computer-aided diagnostic method for multi-class plaque detection and classification. Graphical abstract Diagram of the proposed automatic multi-class coronary atherosclerosis plaque detection and classification framework. ᅟ.

Recent Trends in Artificial Intelligence-Assisted Coronary Atherosclerotic Plaque Characterization

“…As described above, IVUS, IVOCT, and CCTA are the imaging modalities that can be used to characterize coronary atherosclerotic plaques. Research and development of CAD tools for plaque characterization have relied on image datasets from private [ 66 , 67 , 68 ] or public sources [ 69 , 70 , 71 , 72 , 73 ]. Some of the latter are available only upon request [ 25 , 72 ].…”

“…The neighborhood gray tone difference matrix (NGTDM), gray level difference statistics (GLDS), and invariant moment (IM) are incorporated to extract plaque textural features [ 99 ]. In [ 73 ], a multi-class coronary plaque detection framework random radius symmetry (RRS) containing contextual features of plaque was proposed that supplemented the training data of coronary plaques. Likewise, local indicators of spatial association (LISA) and run length (RL) are used to describe the textural features by detection.…”

“…Moreover, RF classifiers can classify the plaque into three classes, namely dense calcium (DC), necrotic core (NC), and fibrotic tissue (FT) and fibro-fatty tissue (FFT). Support vector machines (SVM) [ 73 ] are effectively used for coronary plaque characterization in most studies that utilize different kernel functions. SVM with a second order polynomial kernel function produced better performance parameters for carotid and coronary plaque characterization [ 99 ].…”

Deep learning predicts cardiovascular disease risks from lung cancer screening low dose computed tomography

“…Several machine learning methods were proposed to estimate CVD factors automatically from CT images. A majority of those methods predict clinically relevant image features including CAC scoring 12 – 17 , non-calcified atherosclerotic plaque localization 18 – 22 , and stenosis 23 – 27 from cardiac CT. For LDCT images, subject to motion artifacts and low signal-to-noise ratio in contrast to cardiac CT images, only until recently were deep learning algorithms applied to quantify CAC scoring from LDCT images as a surrogate of the CVD risk 28 – 31 . Even fewer existing methods directly estimate the CVD risk from LDCT images.…”

Deep Learning Predicts Cardiovascular Disease Risks from Lung Cancer Screening Low Dose Computed Tomography

“…Several machine learning methods were proposed to estimate CVD factors automatically from CT images. A majority of those methods predict clinically relevant image features including CAC scoring [13][14][15][16][17] , non-calcified atherosclerotic plaque localization [18][19][20][21][22] , and stenosis [23][24][25][26][27] from cardiac CT. None of the existing methods directly estimates the CVD risk from LDCT images which are subject to motion artifacts and low signal-to-noise ratio in contrast to cardiac CT images. Deep learning algorithms were recently applied to quantify CAC scoring from LDCT images as a surrogate of the CVD risk [28][29][30][31] .…”