The Application Value Of ¹⁸F-FDG PET/CT In Molecular Subtype Classification Of Breast Cancer

Objective:This study aimed to perform a radiomic analysis to evaluate the role of PET/CT imaging in prediction of molecular subtype for breast cancer patients.A total of 273 breast cancer patients with confirmed pathology who all underwent ¹⁸F-FDG PET/CT imaging before any treatment were finally included in this retrospective study.First,PET metabolic parameters were calculated to detect their roles in predicting molecular subtype.Then,multivariate radiomic models were established to detect their predictive performance in molecular subtype classification,and their outperformance over individual PET metabolic parameters was also assessed.In addition,a subgroup analysis involving breast cancer patients whose IHC results for Her-2 were equivocal were also conducted.For this subgroup analysis,first,the status of Her-2 was confirmed by FISH assay,then the included breast cancer patients were divided into Her-2+and Her-2-groups,In the end,a similar radiomic analysis was performed to further verify the predictive power of this radiomic model in Her-2status prediction.Methods:（1）A total of 273 breast cancer patients with confirmed pathology who all underwent ¹⁸F-FDG PET/CT imaging before any treatment from January 1,2010 to December 31,2019 were included in this study;（2）Based on the status of ER,PR and Her-2 by IHC assay or FISH assay,these included breast cancer patients were binary-classified into different groups,namely the Luminal vs Non-Luminal,Her-2+vs Her-2-and TN vs non-TN;（3）Clinical data were collected to be perform comparison between groups for a variety of clinical parameters;（4）All the original image data were obtained and analyzed on AW4.6 workstation to automatically determine the PET metabolic parameters using PETVCAR software,such as SUVmax,SUVmean,SUVpeak,MTV,TLG value.Thus,comparison of these determined PET metabolic parameters were performed in Lunminal vs Non-Luminal,Her-2+vs Her-2-and TN vs non-TN;（5）Image J 1.50i software was used to delineate the ROI of primary breast cancer lesion,then MATLAB software were employed to extract all the radiomic features to calculate Rad-score;（6）Multivariate radiomic models were established to evaluate their performance in prediction of molecular subtype for breast cancer patients by using a ROC analysis.In addition,the AUCs under the ROC were used as the main outcome to reflect the predictive power,and a comparison of the predictive performance in molecular subtype classification was confirmed between the developed radiomic models and individual PET metabolic parameters;（7）A 10-fold cross-validation were performed to verify the predictive power of these established radiomic models in molecular subtype classification for breast cancer patients;（8）A subgroup analysis for Her-2 status prediction involving a total of 154 breast cancer patients was also conducted for those whose IHC results for Her-2 were equivocal in this investigation;（9）The difference of clinical parameters between Her-2+and Her-2-groups was determined in this subgroup analysis;（10）For this subgroup analysis,all the original image data were obtained and analyzed on AW4.6 workstation to automatically determine the PET metabolic parameters using PETVCAR software,such as SUVmax,SUVmean,SUVpeak,MTV,TLG value.Thus,comparison of these determined PET metabolic parameters were performed in Her-2+vs Her-2-groups;（11）3D Slicer 4.10.2 software was used to delineate the ROI of primary breast cancer lesion,then Pyradiomics software in Python 3.7.1 was employed to extract radiomic features and MATLAB software was finally used to select radiomic features and calculate Rad-score;（12）In this subgroup analysis,multivariate radiomic models also established to evaluate its performance in Her-2status prediction by using a ROC analysis.In addition,the AUCs under the ROC were used as the main outcome to reflect the predictive power,and a comparison of the predictive performance in Her-2 status prediction was confirmed between the developed radiomic models and individual PET metabolic parameters;（13）In this subgroup analysis,a 10-fold cross-validation was performed to determine the average performance of this established radiomic model in Her-2 status prediction.Results:（1）The percentage of breast cancer patient with stage I was higher in luminal group than that in other groups（p=0.018）,the incidence of axillary lymph node metastasis was higher in Her-2-group than that in Her-2+group（p=0.048）.There were no obvious difference for the rest of the clinical features between groups;（2）SUVmax,SUVmean and SUVpeak in primary lesion were significantly higher in patients with non-Luminal subtype and TN subtype than that in luminal subtype and non-TN subtype.Whereas no statistically significant difference in those values of SUV were observed between Her-2+and Her-2-groups.（SUVmax,p=0.132;SUVmean,p=0.113;SUVpeak,p=0.121）;（3）No significant difference were found for both MTV and TLG between non-Luminal and Luminal,Her-2+vs Her-2-and TN vs non-TN（p>0.05）;（4）Established multivariate radiomic models significantly outperformed PET metabolic parameter s in predicting molecular subtype of breast cancer for both training cohort and testing cohort,which were reflected with a higher AUC under the ROC for radiomic models that for PET metabolic parameters;（5）The average AUC,accuracy,sensitivity and specificity for developed radiomic models were all higher than 0.8 after a 10-fold cross-validation,except for the specificity in predicting Her-2+vs Her-2-.The average AUC for Luminal vs non-Luminal,Her-2+vs Her-2-and TN vs non-TN were 0.856,0.818 and 0.888,respectively;（6）In a subgroup analysis whose IHC results for Her-2 were equivocal,the status of Her-2was first confirmed by FISH assay.The clinical parameters,such as age,axillary lymph node metastasis and clinical staging were not found with significant difference between Her-2+and Her-2-groups（p>0.05）,but the percentage of mixed carcinoma in Her-2-groups was higher than that in Her-2+group（p=0.046）;（7）In this subgroups analysis,no significant difference between Her-2+and Her-2-groups were not observed for all the five PET metabolic parameters（SUVmax,p=0.111;SUVmean,p=0.104;SUVpeak,p=0.206;MTV,p=0.111;TLG p=0.544）;（8）For this subgroups analysis,developed multivariate radiomic models based on PET/CT images with a higher AUC significantly outperformed PET metabolic parameters in Her-2 status prediction in both training cohort and testing cohort;（9）In this subgroups analysis,the AUC,accuracy,sensitivity and specificity of this built radiomic models for Her-2 status prediction were 0.682,0.706,0.699 and 0.759,respectively.Conclusion:（1）Compared to BC with non-luminal subtype,the radioactive accumulation of ¹⁸F-FDG in primary BC lesions were significantly decreased in BC with luminal subtype.In addition,SUVs,including SUVmax,SUVmean,and SUVpeak were found to be dramatically increased in BC with TN subtype in comparison with that in BC with non-TN subtype.Whereas,SUVs were not able to differentiate between Her-2+and Her-2-BC;（2）In both training and test cohorts,established multivariate radiomic models outperformed single PET metabolic parameter in prediction of molecular subtype for breast cancer;The stability of the predictive performance of these radiomic models in molecular subtype classification for breast cancer was also verified;（3）In a subgroup analysis whose IHC results for Her-2 were ambiguous,PET metabolic parameters were not able to distinguish Her-2+from Her-2-BC.Whereas,established multivariate radiomic models outperformed single PET metabolic parameter in Her-2 status prediction in this subgroup analysis.In addition,the stability of the predictive performance of the developed radiomic models was also verified with great accuracy,sensitivity and specificity;（4）Clinical characteristics such as age,pathological type,axillary lymph node metastasis and staging,and PET metabolic parameters MTV and TLG were not able to predict molecular subtypes of breast cancer,which need to be further explored in future investigation with a larger sample size;（5）In summary,this study proved a significant role of ¹⁸F-FDG PET/CT imaging in molecular subtype classification for breast cancer patients.In particular,established multivariate radiomic model based on PET/CT images exhibited a remarkably predictive power for molecular subtype classification,thus improve individualized diagnosis and treatment for breast cancer patients;（6）This study still have certain shortcomings,which need to be further overcome by prospective and multi-center studies with a larger sample and better research design in future.