student profile: Mr Chongrui Xu


Map

Thesis work

Thesis title: Quantitative radiomics analysis for predictive medicine

Supervisors: Xiu WANG

Thesis abstract:

�p�Radiomics is an emerging field in the field of quantitative standard-of-care images that utilizes high-throughput mining of quantitative image features in standard medical images to enable numerous features to be extracted and applied to clinical decision support systems to improve diagnosis, treatment, and predictive performance. Such medical images, including computed tomography, positron emission tomography or magnetic resonance imaging, extracted features are based on expert delineated region of interest, leading to a gigantic potential database. Radiomic analysis mines complicated image analysis tools and the performance of evaluation for medical image features that uses radiomics signature as predictors to construct precision prediction model providing a powerful tool in modern medicine. There are three different projects in the thesis including, radiomic features in survival prediction for locally advanced non-small-cell lung cancer (LA-NSCLC), 18F-FDG PET / CT for detection of extramedullary lesions in patients with relapsed acute leukemia (AL), and 18F-FDG PET/CT radiomic features to predict pathologic complete response of breast cancer to neoadjuvant chemotherapy, covering the significance of radiomics in the medical field. Importantly, as high dimensionality of the radiomic features available. Therefore, we proposed an integrative clustering and Supervised Feature Selection method based on LA-NSCLC to avoid the problem of overfittings.«/p» «ol style="list-style-type:lower-alpha"» «li style="text-align:justify"»The survival analysis of radiomics: we firstly proposed radiomic predictor for LA-NSCLC patients’ survival analysis. By comparing radiomic features, clinico-pathological and hematological features of 118 cases respectively and adopted cox proportional hazard (CPH) model or random survival forest (RSF).such results as below: radiomic features selected by clustering combined with CPH were found to be more predictive with c-index of 0.699 in comparison to 0.648 by clustering combined with RSF. Based on multivariate CPH model, our integrative nomogram achieved a c-index of 0.792 and retained 0.743 in the cross-validation analysis, outperforming radiomic, clinico-pathological or hematological model alone. The calibration curve of this nomogram showed agreement between predicted and actual values for the 1-year and 2-year survival prediction. Interestingly, the selected important radiomic features were correlated with levels of platelet, platelet/lymphocytes ratio (PLR), and lymphocytes/monocytes ratio (LMR) significantly (p values all <0.05).«/li» «/ol» «p style="text-align:justify"» «/p» «ol start="2" style="list-style-type:lower-alpha"» «li style="text-align:justify"»The diagnosis ability of radiomics: we secondly proposed radiomics improve the diagnostic performance of 18F-FDG PET/CT in detecting bone marrow involvement (BMI) for patients with suspected relapsed AL. Selected 3 representative radiomic features to differentiate BMI for focal uptake, normal uptake, and diffuse uptake patients. Some results as below: The sensitivity, specificity, and accuracy of visual analysis for BMI diagnosis were 62.5%, 73.7%, and 68.6%, respectively. With the cross-validation on the training group, the machine learning model correctly predicted 31 patients in BMI. The sensitivity, specificity, and accuracy of the machine learning model in BMI detection were 87.5%, 89.5%, and 88.6%, respectively, significantly higher than the ones in visual analysis (p < 0.05). The evaluation on the independent validation group showed that the machine learning model could achieve 83.3% accuracy with all the diffuse uptake and normal uptake patients correctly predicted except for one focal uptake patient.«/li» «/ol» «p» «/p» «ol start="3" style="list-style-type:lower-alpha"» «li»The prognostic ability of radiomics: we thirdly proposed according to PET/CT radiomic feature to accurately predict NAC therapeutic efficacy. The training group was composed of 70% patients (70 patients with 35 pCR), and was serving as the base for radiomic feature analysis and used to construct machine learning based prediction model. The rest 30% patients (30 patients with 15 pCR) formed the independent validation group which was utilized to validate the performance of the prediction model. With 30 times random ten-fold cross validation there were 5 PET radiomic features and 3 CT radiomic features finally be selected. The combination pCR prediction model of 2 PET radiomic features with 2 CT radiomic features could achieve the accuracy of 0.857 and 0.767 in the ten-fold validation and the independent validation respectively, outperforming the prediction model purely based on PET radiomic (accuracy=0.714 and 0.7) and CT radiomic (accuracy=0.785 and 0.733). Subgroup analysis of the prediction ability in different molecular types observed that the PET/CT radiomic pattern was more capable of predicting the treatment prognosis for luminal B2 and TN with an accuracy of 0.86 and 0.77, respectively.�/p�

Selected publications

Download citations: PDF RTF Endnote

Journals

  • Wang, L., Dong, T., Xin, B., Xu, C., Guo, M., Zhang, H., Feng, D., Wang, X., Yu, J. (2019). Integrative nomogram of CT imaging, clinical, and hematological features for survival prediction of patients with locally advanced non-small cell lung cancer. European Radiology, 29(6), 2958-2967. [More Information]

Conferences

  • Xin, B., Xu, C., Wang, L., Dong, T., Zheng, C., Wang, X. (2018). Integrative Clustering and Supervised Feature Selection for Clinical Applications. 15th International Conference on Control, Automation, Robotics and Vision (ICARCV 2018), Piscataway: Institute of Electrical and Electronics Engineers (IEEE). [More Information]

2019

  • Wang, L., Dong, T., Xin, B., Xu, C., Guo, M., Zhang, H., Feng, D., Wang, X., Yu, J. (2019). Integrative nomogram of CT imaging, clinical, and hematological features for survival prediction of patients with locally advanced non-small cell lung cancer. European Radiology, 29(6), 2958-2967. [More Information]

2018

  • Xin, B., Xu, C., Wang, L., Dong, T., Zheng, C., Wang, X. (2018). Integrative Clustering and Supervised Feature Selection for Clinical Applications. 15th International Conference on Control, Automation, Robotics and Vision (ICARCV 2018), Piscataway: Institute of Electrical and Electronics Engineers (IEEE). [More Information]

Note: This profile is for a student at the University of Sydney. Views presented here are not necessarily those of the University.