| Lung cancer is the malignancy with the highest incidence(11.6%)and mortality(18.4%)worldwide.Although there is improvement in lung cancer treatment using traditional chemotherapy and antivascular drugs to a certain extent,the 5-year survival rate is still below 20%.The development of targeted therapy has revolutionized patients with lung cancer;however,it is less effective in those with negative driver genes.In recent years,the development of immunotherapy lung cancer treatment has ushered in another leap.The application of immuno-checkpoint inhibitors(ICI)can significantly prolong the progression-free survival(PFS)and overall survival(OS)of patients with advanced lung cancer.However,the individualization of immunotherapy is highly variable,only 20-50%of patients with lung cancer could benefit from ICI treatment.Additionally,the tumor response pattern to immunotherapy is more complex than that of traditional chemoradiotherapy.Besides of the typical response pattern,there are still atypical patterns(such as delayed response,pseudoprogression and hyperprogression).For patients with progression within 8 weeks of initial treatment,neither RECIST1.1 criteria nor PD-L1 monitoring can quickly distinguish between hyperprogression and pseudoprogression.However,the two treatment modalities are inconsistent,and the choice of treatment decision directly affects the efficacy and prognosis of patients.As the only approved biomarker,PD-L1 expression is limited by the spatial and temporal heterogeneity.With the continuous development of artificial intelligence algorithms,radiomic analysis provides an opportunity to extract the micro information of the tumor quantitatively from the macro clinical image,which are challenging to be identified by the naked eye.Currently,radiomics have been successfully applied in various fields of immunotherapy to predict response,prognosis,hyperprogression and pseudoprogression.However,most of the studies research faces the following challenges:First,radiomics models established in most studies only correspond to a single clinical outcome,which cannot meet the needs of complex immunotherapy.Second,the physiological significance of most models remains to be clarified and informed clinical decisions cannot be made.Thirdly,quantitative descriptions of the effects of effective biomarkers other than T cell-mediated antitumor mechanisms of action are lacking.Based on the above deficiencies,we conducted the following research.Part Ⅰ:The prediction of different clinical outcomes of the non-small cell lung cancer patients treated with immunotherapy based on computed tomography(CT)radiomics analysisObjective:To develop a radiomics model based on CT images for predicting different clinical outcomes,aiming to achieve accurate screening of therapeutic benefit population and personalized guidance of clinical decision-making.Materials and Methods:175 patients from our hospital were divided into the training(n=105)and test cohort(n=70)randomly,and another 37 patients from XX hospital were used for external validation.Radiomics features extracted from the primary tumor,peritumoral tumors(5mm,10mm),and lymph nodes were used alone or combined with each other to develop radiomics signatures to predict Durable clinical benefit(DCB),which were further validated for the ability in predicting PFS,OS,typical reactions and atypical reactions.Results:After the comparison of six models,it was found that the combined model RSWPL developed with the primary tumor features,peritumoral tumor features and lymph node features achieve the highest AUCs(area under the curves)in predicting DCB with 0.84,0.82 and 0.81 in the training cohort,test cohort and external test cohort,respectively.It’s noted to say,the incorporation of lymph node features significantly improved the sensitivity of the model to 84.48%,85%and 92.31%in these three cohorts without reducing the specificity.The predicted AUC(area under the curve)of DCB was 0.84,0.82 and 0.81,respectively.As a supplement to PD-L1,the combined model RSWPL can screen an additional 28.57%of PD-L1 negative patients to achieve DCB.RSWPL could be used as an independent prognostic factor to predict PFS and OS(PFS:C-index=0.68,0.66,0.64;OS:C-index=0.61,0.65,0.62).In addition,RSWPL could effectively discriminate patients with typical response(PR/CR,SD,PD),and predict PR/CR patients with the AUC of 0.66,0.71,0.71,respectively.At the same time,it could achieve the prediction of pseudoprogression(AUC=0.80)and the discrimination of hyperprogression(P=0.02).Conclusions:The radiomics models extracted from primary tumors,peritumoral tumors and lymph nodes can be used to predict different clinical outcomes and assist clinicians to make personalized clinical decisions based on the complexity of immunotherapy in a timely and effective manner,which has very practical clinical value.Part Ⅱ:The study of immunotherapy-related molecular markers in animal and clinical experimentsObjective:To find the physiological significance of the radiomics model through animal experiments and clinical patient verification,and to assist clinicians to make informed clinical decision making.Materials and Methods:A total of 30 tumor-bearing mice with the response(LLC)and resistant(LLC-R)immunotherapy were designed.Five mice were selected as the untreated group,and the flow cytometry was performed after tumor formation.Ten mice in each group were randomly divided into the treatment and control groups.After treatment,tumors were collected for flow cytometry.The effective immune cell markers were screened by comparison and immunohistochemistry was performed on patients.The correlation analysis between radiomics model and markers was established to investigate whether it could explain the physiological significance of the above RSWPL model.Results:LLC mice were proved to respond to PD-L1 treatment,while LLC-R mice were not.The subtype of tumor associated macrophage(TAMs)--M2-like TAMs,was the only type of immune cells with the most significant difference between LLC and LLC-R mice before treatment(P=0.016)and the most significant change in LLC mice after treatment(P=0.032).M2-like TAMs were significantly different among patients with different treatment response and prognoses(P=0.029,0.045).RSLymph and RSWPL were independently correlated with M2-like TAMs(coefficients=-0.367,-0.519,P=0.013,<0.001).Conclusion:M2-like TAMs can be used as effective biomarkers for further assessment.The diversified role of M2-like TAMs can effectively explain the physiological basis for the imaging model RSWPL to predict variable clinical outcomes,and the extraction of lymph node features is crucial for this.Part III:The study of noninvasive quantification of immunotherapy efficiency related biomarkers by multimodal radiomicsObjective:To design radiomics models for noninvasive quantification of M2-like TAMs based on multimodal imaging(CT,PET and MRI)which could be further used for prognostic prediction in immunotherapy.Materials and Methods:PET,CT,MR(DWI,T2WI and T2*WI)scans were performed before treatment and before death of the mice.The images of the untreated group were used as the training cohort.The images before and after treatment were used as PT cohort and test cohort,respectively.Results:The prediction performance of M2-like TAMs by multimodal images was good.In the training cohort,AUCs of RSDWI,RST2,RST2STAR,RSCT and RSPET were 0.92,1.00,0.92,1.00 and 0.88,respectively.In the testing cohort,AUCs of RSDWI,RST2,RST2STAR,RSCT and RSPET were 0.82,0.79,0.91,0.88,0.68.AUC of MRS were 1.00 and 0.84 in the training and test cohorts.Multimodal images and combined model can achieve stratified prediction of mice with different response,the former AUCs were 0.80,0.67,0.77,0.72,0.81 and the latter AUC was 0.76.Conclusions:Multimodal radiomics signature can noninvasively quantify M2like TAMs,which has the potential to provide effective screening and monitoring means for new immunotherapies such as macrophage targeted therapy. |
PDF Full Text Request