A Critical Analysis of the Robustness of Radiomics to Variations in Segmentation Methods in 18F-PSMA-1007 PET Images of Patients Affected by Prostate Cancer

BACKGROUND

Radiomics shows promising results in supporting the clinical decision process, and much effort has been put into its standardization, thus leading to the Imaging Biomarker Standardization Initiative (IBSI), that established how radiomics features should be computed. However, radiomics still lacks standardization and many factors, such as segmentation methods, limit study reproducibility and robustness.

AIM

We investigated the impact that three different segmentation methods (manual, thresholding and region growing) have on radiomics features extracted from 18F-PSMA-1007 Positron Emission Tomography (PET) images of 78 patients (43 Low Risk, 35 High Risk). Segmentation was repeated for each patient, thus leading to three datasets of segmentations. Then, feature extraction was performed for each dataset, and 1781 features (107 original, 930 Laplacian of Gaussian (LoG) features, 744 wavelet features) were extracted. Feature robustness and reproducibility were assessed through the intra class correlation coefficient (ICC) to measure agreement between the three segmentation methods. To assess the impact that the three methods had on machine learning models, feature selection was performed through a hybrid descriptive-inferential method, and selected features were given as input to three classifiers, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), Random Forest (RF), AdaBoost and Neural Networks (NN), whose performance in discriminating between low-risk and high-risk patients have been validated through 30 times repeated five-fold cross validation.

CONCLUSIONS

Our study showed that segmentation methods influence radiomics features and that Shape features were the least reproducible (average ICC: 0.27), while GLCM features the most reproducible. Moreover, feature reproducibility changed depending on segmentation type, resulting in 51.18% of LoG features exhibiting excellent reproducibility (range average ICC: 0.68-0.87) and 47.85% of wavelet features exhibiting poor reproducibility that varied between wavelet sub-bands (range average ICC: 0.34-0.80) and resulted in the LLL band showing the highest average ICC (0.80). Finally, model performance showed that region growing led to the highest accuracy (74.49%), improved sensitivity (84.38%) and AUC (79.20%) in contrast with manual segmentation.

PSMA PET Imaging in Glioblastoma: A Preclinical Evaluation and Theranostic Outlook

Background

Prostate specific membrane antigen (PSMA) PET imaging has recently gained attention in glioblastoma (GBM) patients as a potential theranostic target for PSMA radioligand therapy. However, PSMA PET has not yet been established in a murine GBM model. Our goal was to investigate the potential of PSMA PET imaging in the syngeneic GL261 GBM model and to give an outlook regarding the potential of PMSA radioligand therapy in this model.

Methods

We performed an ¹⁸F-PSMA-1007 PET study in the orthotopic GL261 model (n=14 GBM, n=7 sham-operated mice) with imaging at day 4, 8, 11, 15, 18 and 22 post implantation. Time-activity-curves (TAC) were extracted from dynamic PET scans (0-120 min p. i.) in a subset of mice (n=4 GBM, n=3 sham-operated mice) to identify the optimal time frame for image analysis, and standardized-uptake-values (SUV) as well as tumor-to-background ratios (TBR) using contralateral normal brain as background were calculated in all mice. Additionally, computed tomography (CT), ex vivo and in vitro¹⁸F-PSMA-1007 autoradiographies (ARG) were performed.

Results

TAC analysis of GBM mice revealed a plateau of TBR values after 40 min p. i. Therefore, a 30 min time frame between 40-70 min p. i. was chosen for PET quantification. At day 15 and later, GBM mice showed a discernible PSMA PET signal on the inoculation site, with highest TBR_mean in GBM mice at day 18 (7.3 ± 1.3 vs. 1.6 ± 0.3 in shams; p=0.024). Ex vivo ARG confirmed high tracer signal in GBM compared to healthy background (TBR_mean 26.9 ± 10.5 vs. 1.6 ± 0.7 in shams at day 18/22 post implantation; p=0.002). However, absolute uptake values in the GL261 tumor remained low (e.g., SUV_mean 0.21 ± 0.04 g/ml at day 18) resulting in low ratios compared to dose-relevant organs (e.g., mean tumor-to-kidney ratio 1.5E^-2 ± 0.5E^-2).

Conclusions

Although ¹⁸F-PSMA-1007 PET imaging of GL261 tumor-bearing mice is feasible and resulted in high TBRs, absolute tumoral uptake values remained low and hint to limited applicability of the GL261 model for PSMA-directed therapy studies. Further investigations are warranted to identify suitable models for preclinical evaluation of PSMA-targeted theranostic approaches in GBM.

18F-PSMA-1007 PET/CT for response assessment in patients with metastatic renal cell carcinoma undergoing tyrosine kinase or checkpoint inhibitor therapy: preliminary results

Introduction

Tyrosine kinase (TKI) and checkpoint inhibitors (CI) prolonged overall survival in metastatic renal cell carcinoma (mRCC). Early prediction of treatment response is highly desirable for the individualization of patient management and improvement of therapeutic outcome; however, serum biochemistry is unable to predict therapeutic efficacy. Therefore, we compared ¹⁸F-PSMA-1007 PET imaging for response assessment in mRCC patients undergoing TKI or CI therapy compared to CT-based response assessment as the current imaging reference standard.

Methods

¹⁸F-PSMA-1007 PET/CT was performed in mRCC patients prior to initiation of systemic treatment and 8 weeks after therapy initiation. Treatment response was evaluated separately on ¹⁸F-PSMA-PET and CT. Changes on PSMA-PET (SUV_mean) were assessed on a per patient basis using a modified PERCIST scoring system. Complete response (CR_PET) was defined as absence of any uptake in all target lesions on posttreatment PET. Partial response (PR_PET) was defined as decrease in summed SUV_mean of > 30%. The appearance of new, PET-positive lesions or an increase in summed SUV_mean of > 30% was defined as progressive disease (PD_PET). A change in summed SUV_mean of ± 30% defined stable disease (SD_PET). RECIST 1.1 criteria were used for response assessment on CT. Results of radiographic response assessment on PSMA-PET and CT were compared.

Results

Overall, 11 mRCC patients undergoing systemic treatment were included. At baseline PSMA-PET₁, all mRCC patients showed at least one PSMA-avid lesion. On follow-up PET₂, 3 patients showed CR_PET, 3 PR_PET, 4 SD_PET, and 1 PD_PET. According to RECIST 1.1, 1 patient showed PR_CT, 9 SD_CT, and 1 PD_CT. Overall, concordant classifications were found in only 2 cases (2 SD_CT + PET). Patients with CR_PET on PET were classified as 3 SD_CT on CT using RECIST 1.1. By contrast, the patient classified as PR_CT on CT showed PSMA uptake without major changes during therapy (SD_PET). However, among 9 patients with SD_CT on CT, 3 were classified as CR_PET, 3 as PR_PET, 1 as PD_PET, and only 2 as SD_PET on PSMA-PET.

Conclusion

On PSMA-PET, heterogeneous courses were observed during systemic treatment in mRCC patients with highly diverging results compared to RECIST 1.1. In the light of missing biomarkers for early response assessment, PSMA-PET might allow more precise response assessment to systemic treatment, especially in patients classified as SD on CT.