| Objection in all:The development of radiomics is currently tending towards the inclusion of increasingly diverse sources of information,including imaging methods,imaging sequences,clinical and biochemical examinations,among others.The large volume of data involved places ever-increasing demands on the hardware and software used in data processing,which has been one of the factors hindering the wider clinical application of radiomics.In this study,a multi-center design with external validation was employed to investigate the diagnostic performance of seven radiomics features,ranging from simple to complex,based on two levels of microvascular invasion(MVI)in tissue pathology and the expression of the biomarker Ki-67.The study aimed to demonstrate the superiority of multi-parameter radiomics features and the feasibility of simple radiomics features,in order to provide objective,non-invasive,and efficient radiological evidence for the busy and demanding clinical imaging diagnosis work.The study also aimed to provide more objective,accurate,and non-invasive radiological supplementary materials for primary medical institutions that have not yet conducted enhanced scanning.This study attempted to construct a predictive model for postoperative tumor recurrence time and perform high and low recurrence risk stratification prediction by using the aforementioned best combination of radiomics features.Through the correlation analysis of MVI,Ki-67,and recurrence time,the study preliminarily revealed the relationship between MVI,Ki-67,and tumor recurrence time.Chapter 1:Clinical application of multi-parameter plain scan MRI Radiomics in predicting MVIObjective:A multi-center study was conducted to investigate the clinical application value of multi-parameter non-enhanced MRI radiomics features for predicting the occurrence of microvascular invasion(MVI)in hepatocellular carcinoma(HCC)lesions.By comparing the predictive performance of various combinations of multi-parameter non-enhanced MRI radiomics features,the study aimed to demonstrate the accuracy and superiority of multi-parameter radiomics features,as well as the feasibility of simple radiomics features.Materials and Methods:Retrospective data of 209 HCC patients,including 127 cases from the author’s unit and 82 cases from three external centers(34+30+18 cases),were collected.All patients underwent surgery to remove the tumor,were pathologically confirmed as HCC,and completed MVI evaluation.We collected the clinical data of these patients and MRI data within 2 months before surgery.The AK software was used to manually delineate all lesion areas on T1WI,T2WI,and DWI images,and feature data were extracted from each region of interest(ROI)of each lesion.A total of 851 feature data were extracted from each sequence,and 2,553 feature data were extracted from each patient.161 data from this center and one of the external centers were used as the training group,and the remaining 48 data from the other two external centers were used as the validation group.A multi-parameter imaging biomarker prediction model was constructed using feature selection and Logistic regression.To explore the accuracy and superiority of multi-parameter imaging biomarkers,seven modeling methods were used:three sequence features(T1WI,T2WI,DWI)/any two sequence combinations/any one sequence feature data(T1WI+T2WI+DWI,T1WI+T2WI,T1WI+DWI,T2WI+DWI,T1WI,T2 WI,DWI).The efficacy of each model in predicting MVI was evaluated using the area under the receiver operating characteristic curve(AUC)and calibration curve.Results:The DWI+T1WI+T2WI,T1WI+T2WI,DWI+T2WI,DWI+T1WI,T1WI,T2WI,DWI imaging biomarker models in the training group had AUCs of 0.827,0.763,0.749,0.832,0.745,0.706,and 0.731 for predicting MVI,respectively,all with P values less than 0.05.The AUCs of the validation group were 0.837,0.737,0.772,0.833,0.722,0.748,and 0.717,respectively,all with P values less than 0.05.Conclusion:Based on non-enhanced MRI multi-parametric imaging biomarkers,MVI can be accurately predicted.The performance of multi-parametric imaging biomarkers using three sequences is similar to that of the DWI+T1WI combined model for predicting MVI.However,the advantages of multi-parametric imaging biomarkers using the three sequences are more significant compared to single sequence or other two-sequence model imaging biomarkers.Chapter 2:A Preliminary Study of Multi-Parametric Plain MRI Image-Based Radiomics Labels for Predicting Ki-67 Levels in Hepatocellular Carcinoma.Objective:To explore the clinical application value of multi-parametric MRI radiomics features for predicting Ki-67 expression level in HCC lesions based on a multicenter study using non-enhanced MRI.By comparing the predictive performance of various combinations of multi-parameter non-enhanced MRI radiomics features,the study aimed to demonstrate the accuracy and superiority of multi-parameter radiomics features,as well as the feasibility of simple radiomics features.Materials and Methods:Clinical data and seven modalities of radiomics features were collected from a total of 129 HCC patients confirmed by pathological diagnosis and Ki-67 evaluation,including 56 cases from the author’s unit and 73 cases from three other centers(30+29+14 cases).Ki-67≤20%was defined as low level,and Ki-67>20%was defined as high level.Seven radiomics models were constructed to predict Ki-67 expression level,using the data from the author’s unit and one external center as the training group,and the data from the other two external centers as the validation group.ROC-AUC and calibration curves were used to evaluate the predictive performance of each model.Results:In the training group,the DWI+T1WI+T2WI,T1WI+T2WI,DWI+T2WI,DWI+T1WI,T1 WI,T2WI,and DWI radiomics models had AUCs of 0.862,0.790,0.801,0.739,0.787,0.746,and 0.720,respectively,in predicting Ki-67 expression level.In the validation group,the corresponding AUCs were 0.798,0.663,0.665,0.717,0.710,0.647,and 0.669.Conclusion:Based on MRI multi-parametric radiomics labels,it is possible to accurately predict the expression level of Ki-67.Three-sequence radiomics label models have better classification abilities than two-sequence or single-sequence models,but each two-sequence model does not show better classification performance than the single-sequence model.Combining the results of the first part of the study,it fully indicates that more sequences do not necessarily mean better performance in multi-parametric radiomics research,and increasing the amount of feature data also comes with a certain degree of workload consumption.With the continuous optimization of imaging technology and further development of radiomics algorithms,multi-parametric radiomics labels may face greater challenges.Balancing the increase in workload and the increase in AUC in clinical work requires careful consideration.Chapter 3:A preliminary study on predicting the time of postoperative recurrence using multi-parametric MRI radiomics features.Objective:To use a multicenter study to explore the use of plain MRI multi-parameter radiomics labels to predict the time of postoperative recurrence in HCC.Materials and Methods:The imaging diagnostic system and clinical follow-up data of the first part of the enrolled patients were retrieved.Patients without any postoperative follow-up records were excluded from the group.The first recorded time of tumor recurrence or the most recent imaging examination indicating no recurrence was recorded.A total of 163 HCC patient data were included in the study,with 90 cases from the author’s unit and 73 cases from three external centers(26+30+17 cases).All included cases were divided into a training group of 130 cases and a testing group of 33 cases in an 8:2 ratio.Three sequence radiomics feature data were used,and a risk model for recurrence was constructed using COX regression method after feature selection.Patients were classified into high and low-risk groups based on the median value of their recurrence risk calculated by the model.KM curve analysis was used to show the classification ability of the model.Results:A three-sequence radiomic model predicted postoperative recurrence with a C-Index of 0.688.The average C-Index value for the five-fold cross-validation was 0.6693.The median time to recurrence for patients was 23 months,and patients were divided into low-risk and high-risk groups based on a cutoff of 24 months.The survival KM curve suggested that the model could predict high and low recurrence risks well(p=0.0019).Conclusion:imaging radiomics labels based on multi-parameter plain MRI can provide an initial estimation of the time of postoperative recurrence and can predict the stratification of high and low recurrence risks well.Chapter 4:The Relationship between MVI and Ki-67 and Postoperative recurrence timeObjective:To study the relationship between the occurrence of MVI and Ki-67 levels within HCC lesions and Postoperative recurrence time.Materials and Methods:MVI and Ki-67 data were extracted from 129 patients in Part 2 and analyzed for the relationship between MVI occurrence and Ki-67 levels.Time data on MVI and tumor recurrence were extracted from 163 patients in Part 3 to analyze the relationship between MVI occurrence and tumor recurrence time.The MVI and Ki-67 data and tumor recurrence time data of 113 patients who completed MVI and Ki-67 evaluation and tumor follow-up were extracted from the second and third parts.,and analyzed for the relationship between Ki-67 levels and tumor recurrence.The relationship between MVI,Ki-67,and tumor recurrence was comprehensively analyzed.Results:The AUC for predicting MVI occurrence with high Ki-67 levels was 0.765(95%CI 0.683-0.848).The KM curve indicated a significant difference in recurrence risk between MVI-negative and MVI-positive patients(p=7.573×10-7),with MVI-positive patients having shorter recurrence times.The expression of Ki-67 at different levels showed a significant difference in survival time(P=0.006),with high expression leading to shorter recurrence times.The combined use of MVI and Ki-67 had a stronger correlation with tumor recurrence one year after surgery compared to using either indicator alone,with r values of 0.334,0.297,and 0.269,respectively.Conclusion:There is a high correlation between high expression of Ki-67 and the occurrence of MVI;lesions with MVI-positive status have a shorter recurrence time,and lesions with high expression of Ki-67 also have a shorter recurrence time.The simultaneous presence of MVI and high Ki-67 levels is more closely related to tumor recurrence within one year after surgery,indicating that patients with MVI-positive status and high Ki-67 expression should be given more attention and a closer follow-up review process.Summary:The research team found that using MRI-based multi-parameter radiomics signatures could accurately predict the occurrence of MVI and Ki-67 levels,and provide some indication for predicting the recurrence of HCC within 24 months after surgery.The three-sequence radiomics signature model showed slightly better performance than the two-sequence or single-sequence models,and the classification performance of the two-sequence models was not significantly different from that of the single-sequence models,but the increased workload of feature data extraction also resulted in some consumption.It is not always true that more parameters and image feature data lead to better predictive performance in multi-parameter radiomics research.With the continuous improvement of imaging technology and the further development of radiomics algorithms,simple image combination models may have comparable classification ability,and multi-parameter/multi-modal radiomics research may face greater challenges in balancing the workload and AUC increase in clinical practice.The occurrence of MVI and high Ki-67 levels at the same time is more closely related to tumor recurrence within one year after surgery.This indicates that patients with high Ki-67 expression and MVI positivity should be given more attention and closely monitored during follow-up examinations. |
