Application Of A Machine Learning Model Based On Radiomics Features In Assessing Carotid Plaque Stability And A Real-world Study Of Drug-coated Balloons In Chronic Occlusive Disease Of The Femoral Popliteal Artery

Machine Learning-Derived Radiomic Signature of Carotid Atherosclerosis Plaque using Computed Tomography Angiography to Evaluate Plaque StabilityPurpose:About one-third of strokes are caused by carotid artery stenosis.When a carotid artery plaque ruptures,it will cause acute embolism,and trigger cerebral ischemia.Compared with stable plaques,unstable plaques have a higher chance of rupture and a greater risk of stroke.A larger number of studies have pointed out that plaque stability is an important factor of prognosis and can also guide the choice of open surgeries or endovascular treatment.However,limited by the ability of human eyes and subjective judgment,the current non-invasive imaging technology can only distinguish few morphological characteristics of plaques and cannot observe most microscopic features.But radiomics can automatically and quantitatively extract many high-dimensional texture features that is difficult to recognize by the human eye through computer post-processing technology.Therefore,this study tried to use imaging data based on computed tomography angiography(CTA),combined with the histological results of carotid artery atherosclerotic plaque to explore whether radiomics and machine learning can evaluate the stability of carotid artery plaque well.Methods:This study is a single-center retrospective research including 68 patients between December 2014 and August 2019.All patients were performed CTA and all plaque samples were saved.After the segmentation of CTA images of carotid plaque was performed semi-automatically,radiomic features were extracted with PyRadiomics modules based on Python.And the principal components were extracted through principal component analysis.Hematoxylin-eosin(HE)staining is used in all plaque samples and plaques were divided into stable and unstable groups with AHA classification,size of lipid-rich necrotic cores,thickness of fibrous caps and calcification.Finally,random forest models were established based on plaque stability.The training set and the test set are randomly divided according to 7:3,and the training set uses a five-fold cross-validation method to train the model,and the above process is repeated 5 times to ensure the robustness of the results.The effect of the model was evaluated by the ROC analysis results of the test set.And Mann-Whitney’s U test and ROC analysis between groups were also performed on the original radiomic features.Results:68 patients received carotid endarterectomy were enrolled,including 39(57.4%)in the stable plaque group and 29(42.6%)in the unstable group.No significant difference exists in clinical information,plaque morphology and plaque pathology between the two groups,but the mean Hounsfield unit of unstable group is higher.In the principal component analysis of the radiomic features,this study retained 90%of the interpretable variance,and extracted 18 principal components of radiomic features,and the Bonferroni adjusted p value was 0.0009 when testing hypotheses of original radiomic features.After repeating the procedure of random forest 5 times for better robustness,the AUC of the clinical data group was 0.55(95%CI:0.44-0.66,p=0.438),while the AUC of the radiomic group and the combination group were 0.68(95%CI:0.57-0.79,p<0.001)and 0.69(95%CI:0.59-0.79,p=0.001).The effect of the model is significantly improved after the radiomic features are included(p=0.013),but whether the clinical data is included or not has no significant effect.In the importance of the characteristics of the combined group model,it is seen that the principal components of radiomic are the main part,and the low-density lipoprotein and the maximum systolic blood pressure also have a certain effect on the model.In the analysis of the original radiomic features,there are 72 features(4.9%)that have significant differences(comparing to adjusted p value 0.0009)between groups.Conclusion:This study proved that radiomics can improve the preoperative evaluation of plaque stability to some extent through the routine CTA examination.In the future,we will increase sample size,combine more types of imaging data,and optimize the machine-learning algorithm to improve the performance of prediction models and stride towards precision medicine.Single-Center Real-World Research of Drug-Coated Balloons for Long and Occlusive Femoropopliteal Artery DiseasePurpose:Drug-coated balloons(DCBs)has been widely used in femoropopli teal lesions.But most related trials focus in short lesions and cohort with relatively low proportion of chronic total occlusion(CTO)lesions.However,there is limited evidence of DCBs in femoropopliteal CTO lesions,especially long and CTO lesions which are not rare in clinical work.Thus,this study is designed to investigate the efficacy and safety of the DCBs for treatment of long and CTO femoropopliteal arterial lesions.Methods:This study is a retrospective single-arm real-world trial,including 44 lesions from 44 patients with the average follow-up period of 12.76±7.13 months.Lesions are CTO lesions and longer than 10cm and are treated with endovascular treatment using DCBs.Endpoints includes primary patency,target lesion revascularization(TLR),the change of Rutherford classification,amelioration of ankle brachial index(ABI)and major adverse events(defined as all-cause death and major target limb amputation).Results:In this study,technical success was 97.7%.The average lesion length was 186±86.3mm and 13.6%of patients received stent implantation.The one-year cumulative primary patency was 78.8%±6.8%,and the one-year cumulative TLR rate was 91.4%±4.9%.The Rutherford classification changed from 3.3±1.0 before operation to 2.1±1.4 at follow-up(p<0.001).The amelioration rate was 72.7%.The ABI changed from 0.33±0.40 to 0.67±0.37(p=0.002).Any DCB-related adverse event was not found.In multivariate analysis which is to explore independent prognostic factors,the significant results are hypertension and restenosis.Conclusion:The study suggests that it is effective and safe to use DCBs to treat femoropopliteal long CTO lesions,at least during one year after the procedure.Most lesions can keep patent and no DCB-related adverse event was observed.But studies with larger scale in the future are still needed to confirm this conclusion.Single-Center Real-World Research of Drug-Coated Balloons for Long and Occlusive Femoropopliteal Artery DiseasePurpose:Drug-coated balloons(DCBs)has been widely used in femoropopliteal lesions.But most related trials focus in short lesions and cohort with relatively low proportion of chronic total occlusion(CTO)lesions.However,there is limited evidence of DCBs in femoropopliteal CTO lesions,especially long and CTO lesions which are not rare in clinical work.Thus,this study is designed to investigate the efficacy and safety of the DCBs for treatment of long and CTO femoropopliteal arterial lesions.Methods:This study is a multi-center retrospective single-arm real-world trial,including data from Peking Union Medical College Hospital,China-Japan Friendship Hospital and Peking University First Hospital.There are 150 lesions from 145 patients with the average follow-up period of 12.55±6.62 months.Lesions in both parts are CTO lesions and longer than 10cm and are treated with endovascular treatment using DCBs.Endpoints includes primary patency,target lesion revascularization(TLR),the change of Rutherford classification,amelioration of ankle brachial index(ABI)and major adverse events(defined as all-cause death and major target limb amputation).Results:In this study,technical success was 98.0%.The average lesion length was 230±82.4nm and 24.6%of patients received stent implantation.The one-year cumulative primary patency was 80.9%±3.9%,and the one-year cumulative TLR rate was 84.7%±3.8%.The Rutherford classification changed from 3.5±1.0 before operation to 1.5±1.4 at follow-up(p<0.001).The ABI changed from 0.49±0.27 to 0.87±0.35(p<0.001).Six patients(4.14%)were found dead in follow-up due to non-cardiovascular diseases,regarded as non-procedure related death.In multivariate analysis which is to explore independent prognostic factors,the significant results are renal dysfunction and non-in-stent restenosis.Conclusion:The study suggests that it is effective and safe to use DCBs to treat femoropopliteal long CTO lesions,at least during one year after the procedure.Most lesions can keep patent and no DCB-related adverse event was observed.But studies with larger scale are still needed to confirm the conclusions.