Development Of A Machine Learning-based Radiomics Signature For Estimating Breast Cancer TME Phenotypes And Predicting Anti-PD-1/PD-L1 Immunotherapy Response

Background:Breast cancer is one of the common malignancies in women,and the incidence rate ranks first among female malignant tumors,which seriously endangers women’s health.Immunotherapy,represented by PD-L1 inhibitors,is an emerging approach currently for breast cancer treatment.Yet,the response to immunotherapy varies greatly between individuals.Accurately predicting the success of immunotherapy using biomarkers generated from the tumor microenvironment is hampered by the invasive nature of obtaining genetic information.Radiomics offers numerous insights on the diagnosis and prognosis of breast cancer due to its non-invasive examination of enormous amounts of image data.Consequently,the goal of this work was to create and independently verify a radiomic biomarkers based on the breast cancer tumor microenvironment phenotype.Methods:In this retrospective multi-cohort study,this thesis focuses on 1)calculate the extent of immune infiltration in the tumor microenvironment of breast cancer by RNA-seq data and assess it using radiomics;2)by using a bioinformatics approach,develop a tumor microenvironment phenotype of breast cancer that elucidate the complex cellular interactions and infiltration patterns;analyze the correlation between the microenvironmental phenotypes and prognosis of breast cancer;3)explore vast amount of information in DCE-MRI by radiomic and construct radiomic signatures for non-invasively and longitudinally evaluating the tumor microenvironment phenotype of breast cancer followed by internal and external validation of the signatures;4)further predict the clinical response of patients receiving anti-PD-1/PD-L1 immunotherapy by the radiomic signatures with a generalisation validation.Results:The degree of immune cell infiltration in breast cancer was found to be strongly correlated with patient prognosis(P<0.05),and the tumor microenvironment could be classified into two distinct phenotypes: immune inflammation and immune desert.The radiomic can predict the tumor microenvironment phenotype of breast cancer(area under the curve([AUC]=0.855;95% CI 0.777-0.932;P<0.05)and has been validated by an internal validation cohort(0.844;0.606-1;P<0.05).In the independent external validation cohort,the signatures also distinguished between two phenotypes(0.814;0.717-0.911;P<0.05).Patients in the immunotherapy cohort had higher radiomic scores than those with the tstable or progressive disease(P<0.05);meanwhile,radiomic signatures had good prediction performance for immunotherapy response(0.784;0.643-0.926;P<0.05).Conclusion:This study develops a radiomic signature to evaluate the breast cancer tumour microenvironment phenotype,predict the efficacy of anti-PD-1/PD-L1 immunotherapy patients,screen the patient population suitable for immunotherapy,and aid in the decision-making process for individualized immunotherapy for breast cancer.