Using Multi-modality Medical Images And Patient-specific Biomechanical Models To Predict Coronary Plaque Progression And Vulnerability

Atherosclerotic plaques may rupture without warning leading to subsequent thrombosis and acute cardiovascular syndromes.In recently years,thin fibrous cap(less than 65 μm)of a vulnerable plaque has become one of the closely monitored risk factors in assessing plaque progression and vulnerability(defined as the likelihood of plaque may rupture or critical events may occur in the near future).Due to its resolution limitation(normally 100-200 μm),intravascular ultrasound(IVUS)medical imaging cannot identify vulnerable plaques whose fibrous cap thickness is less than its resolution,nor can it provide accurate and reliable plaque cap thickness data,which directly affects assessment of plaque status and vulnerability.With its superior resolution(normally 10-15 μm),optical coherence tomography(OCT)has made accurate quantification of coronary plaque fibrous cap thickness possible.Combining the advantages of deep penetration from IVUS and high resolution of OCT,IVUS,OCT,coronary angiography and clinical data of 10 patients with coronary artery disease at baseline and oneyear follow-up were used in this study.Combining the advantages of deep penetration from IVUS and high resolution of OCT,image segmentation and fusion were performed to generate multi-component segmentation data of IVUS and multi-modality IVUS+OCT(IO)data,and patient-specific biomechanical models were constructed to compared the differences of plaque morphology and stress and strain between IVUS and IO.Generalized linear mixed model(GLMM),least squares support vector machine(LSSVM)and random forest machine learning methods were used to predict plaque cap thickness and vulnerability changes.Clear prediction accuracy improvements using IO data and models were observed.Results are given below.Accurate quantification of plaque morphology and fibrous cap thickness.Quantitative cap thickness results from IVUS and IO data showed that the relative differences range of minimum fibrous cap thickness between IVUS and IO data was ? 91.2%～69.6%,and the relative difference range of average fibrous cap thickness was ? 44.5%～73.9%.The comparisons of other common morphological parameters showed similar results.These results demonstrated quantitatively the possible errors by IVUS data.Multi-modality approach could improve the accuracy of plaque morphology and fibrous cap thickness data.Accurate calculation of plaque stress and strain.114 matched slices(IVUS and IO matching,baseline and follow-up matching)were selected to construct 3D thin-wall models.Stress and strain distributions at fibrous cap from both IVUS and IO models were extracted and compared.The results showed that the relative difference range of maximum cap stresses bewteen IO and IVUS models was ? 34.7% ～ 87.4%.The relative difference range of maximum cap strain was ? 25% ～ 12.4%.These preliminary results showed the importance of high resolution multi-modality data for plaque stress and strain calculations.Improve prediction accuract for fibrous cap thickness change.Choosing cap thickness as the measure for plaque change and using nine morphological and mechanical risk factors to predict fibrous cap thickness change,the results showed that the prediction accuracy of the best combination predictor using GLMM and IO data was significantly higher than that using IVUS data(90.8% vs.74.6%).Using the LSSVM method,the prediction accuracy of the best combination predictor based on IO data was also higher than that using IVUS data(75.7% vs.69.6%).Improve plaque vulnerability change prediction accuracy.Based on accurate multimodality baseline and follow-up data,minimum fibrousc cap thickness,mean fibrous cap stress and mean fibrous cap strain vulnerability indices were defined to perform plaque vulnerability prediction research.The best prediction accuracies for cap thickness index,cap stress index,and cap strain index using combined morphological and mechanical predictors were 90.3%(AUC = 0.88),85.6%(AUC = 0.87)and 83.3%(AUC = 0.81),respectively.The prediction accuracy for cap stress index of the best combination predictor was 10% better than that from the best single predictor(85.6% vs.75.6%).Prediction accuccy of the stress index using IO data was 16.9% better than that using IVUS data(85.6% vs.68.7%).Above results have important clinical importance for disease monitoring and patient management.The innovations of this paper include: 1)Accurate quantification of fibrous cap thickness and plaque morphology using IVUS+OCT multi-modality image data.Multimodality IO data improved the accuracy and reliability of fibrous cap thickness measurements.2)Accurate plaque morphologies and biomechanical models based on IO data provided accurate and reliable plaque stress/strain calculations with which relative error ranges for stress/strain data from IVUS models were determined.3)With accurate multi-modality IO cap thickness and plaque stress/strain data,morphological and mechanical risk factors were combined to improve the accuracy and reliability of plaque cap thickness change predictions.4)Multi-modality image and plaque stress/strain data provided more accurate plaque vulnerability predictions.The differences in the accuracies of predicting vulnerability changes between IVUS and IO data were compared.