Clinical Assistant Diagnosis Research Based On Pattern Recognition And Statistical Analysis

In recent years,quantitative medical image processing has been increasingly used in the clinical assistant diagnosis research.The high-dimensional radiographic information of tumor heterogeneity or microenvironment extracted from medical images can be used to construct the clinical assistant diagnosis models,which can help the doctor to cope with the heavy clinical work and even possibly improve the clinical diagnosis accuracy.In this paper,we employ pattern recognition and statistical analysis to conduct clinical assistant diagnosis research.(1)Quantitative assessment of pathological good response to neoadjuvant chemoradiation for locally advanced rectal cancer based on difference in difference regression analysisAccurate prediction of the tumor response of neoadjuvant chemoradiation for locally advanced rectal cancer is vital for making the surgical decision of organ-preserving or total resection strategies.We extract high-dimensional quantitative features from diffusion weighted images(DWI)and use difference in difference regression analysis to select the features predictive of tumor response.Then,elastic-net logistic regression model is used to construct the DWI prediction model based on the selected features.In addition,we also construct the clinical prediction model and the combined prediction model based on the clinical characteristics and DWI signature.We find out that the DWI prediction model obtains high positive predictive values and the combined prediction model achieves satisfying overall prediction performance.It indicates that the quantitative analysis of DWI data combined difference in difference regression analysis is potent in prediction of good response to neoadjuvant chemoradiation,which might help to provide decision operative support for locally advanced rectal cancer patients.(2)Multi-scale quantitative enhanced-CT analysis based prediction of benign and malignant renal massesClinically,it is difficult to differentiate fat-poor angiomyolipoma(fp-AML)and renal cell carcinomas(RCC).In this study,we conduct multi-scale quantitative analysis on enhanced-CT images and extract three kinds of high-dimensional features,including including histogram-based features,texture-based features and laws’ features.We obtain seven different feature combination sets from the three kinds of features,and make prediction of fp-AML and RCC.We find that the histogram based features combined with texture-based features and laws’ features are efficient in prediction of fp-AML and RCC,achieving the mean prediction accuracy of 91.81%.Nevertheless,the different scales don’t have obvious effect on the prediction performance.(3)Detection of AIDS-related White Matter Injuries based on Tract-based spatial statistical analysis and Multivariable Pattern AnalysisHuman immunodeficiency virus(HIV)infection would lead to cognitive impairment of the patients.White matter damage is considered as vital in the process of cognitive impairment.Due the difficulty of conducting longitudinal research in the clinic,we employ the simian immunodeficiency virus(SIV)affected rhesus monkey model to explore the longitudinal white matter damage.We find that the SIV infection will lead white matter impairments in the inferotemporal regions,which are also significantly correlated with immunoreaction.We employ the multivariate pattern analysis to investigate the HIV-related changes in white matter integrity and connections.We find that the white matter integrity in the descending motor pathway is impaired and the structure connections to the prefrontal regions is disrupted due to HIV infection.