Background: Prostate cancer (PC) is a leading cause of cancer-related deaths in men worldwide. PSMA-directed positron emission tomography (PET) has shown promising results in detecting recurrent PC and metastasis, improving the accuracy of diagnosis and treatment planning. To evaluate an artificial intelligence (AI) model based on [18F]-prostate specific membrane antigen (PSMA)-1007 PET datasets for the detection of local recurrence in patients with prostate cancer. Methods: We retrospectively analyzed 1404 [18F]-PSMA-1007 PET/CTs from patients with histologically confirmed prostate cancer. Artificial neural networks were trained to recognize the presence of local recurrence based on the PET data. First, the hyperparameters were optimized for an initial model (model A). Subsequently, the bladder was localized using an already published model and a model (model B) was trained only on a 20 cm cube around the bladder. Finally, two separate models were trained on the same section depending on the prostatectomy status (model C (post-prostatectomy) and model D (non-operated)). Results: Model A achieved an accuracy of 56% on the validation data. By restricting the region to the area around the bladder, Model B achieved a validation accuracy of 71%. When validating the specialized models according to prostatectomy status, model C achieved an accuracy of 77% and model D an accuracy of 77%. All models achieved accuracies of almost 100% on the training data, indicating overfitting. Conclusions: For the presented task, 1404 examinations were insufficient to reach an accuracy of over 90% even when employing data augmentation, including additional metadata and performing automated hyperparameter optimization. The low F1-score and AUC values indicate that none of the presented models produce reliable results. However, we will facilitate future research and the development of better models by openly sharing our source code and all pre-trained models for transfer learning.
Prostate cancer (PC) is a leading cause of cancer-related deaths in men worldwide. PSMA-directed positron emission tomography (PET) has shown promising results in detecting recurrent PC and metastasis. We developed a deep-learning model to predict cancer recurrence from PSMA PET/CT images of prostate cancer patients after treatment. Different methods were used to improve the performance of the initial model, such as modifying the region of interest (ROI), including metadata as additional layers, or passing prostatectomy state to the model. A hyperparameter optimization of multiple parameters was performed to further increase the model's performance, which, combined with including metadata as additional image layers, resulted in the best validation accuracy of 77 %. Even though the final validation accuracy fell short of reaching 90 %, significant improvements were made to the model, and many approaches were tested that can be a basis for further research and improvements toward the performance of a model that can confidently detect cancer remission in the prostate or prostate bed.
Background: Quantitative dynamic contrast-enhanced myocardial perfusion cardiovascular magnetic resonance (CMR) enables estimation of myocardial blood flow (MBF) and myocardial perfusion reserve (MPR). These measurements require an arterial input function (AIF), which is typically derived from the left ventricular blood pool. However, the contrast agent bolus undergoes dispersion during transport through the coronary vasculature before reaching the myocardial microcirculation. This may introduce systematic and spatially heterogeneous errors in MBF and MPR estimates. Purpose: This work provides an extended segmental analysis of bolus-dispersion-induced errors in quantitative myocardial perfusion MRI using previously established computational fluid dynamics (CFD) simulations in realistic porcine coronary artery models. The focus of the present analysis is the assignment of coronary outlets to myocardial segments and the resulting segmental variability of MBF and MPR errors. Methods: Realistic three-dimensional models of the left and right coronary arteries were extracted from an ex-vivo porcine imaging cryomicrotome dataset. The models extended down to the pre-arteriolar level and included 364 outlets for the left coronary artery and 104 outlets for the right coronary artery, with an average outlet diameter of 383 ± 85 µm. Blood flow was simulated under rest and stress conditions using OpenFOAM. Contrast agent transport was then modeled by solving the advection-diffusion equation using a gamma-variate bolus as input. Outlet concentration-time curves were analyzed using an indicator-dilution model to estimate MBF and MPR errors. Outlets were assigned to standardized myocardial segments, and segmental averages were evaluated with respect to coronary supply territory and travel distance from the model inlet. Results: The simulations demonstrated marked segmental heterogeneity of volume blood flow and bolus-dispersion-induced MBF and MPR errors. Errors increased with travel distance from the coronary artery inlet and were more pronounced in regions supplied by the right coronary artery, consistent with lower flow velocities and stronger bolus dispersion. The resulting systematic errors led to underestimation of MBF and overestimation of MPR, with segmental deviations reaching up to approximately 60%. Conclusion: Bolus dispersion in the coronary vasculature may lead to substantial segmental and location-dependent errors in quantitative myocardial perfusion MRI. This extended analysis indicates that dispersion-related bias is not spatially uniform, but depends on coronary supply territory, travel distance, and flow conditions. These effects should be considered when interpreting regional MBF and MPR estimates, particularly as automated quantitative myocardial perfusion CMR becomes more widely used.
Objective: On dual-tracer positron emission tomography/computed tomography (PET/CT) with 2-deoxy-2-[¹⁸F]fluoro-D-glucose ([¹⁸F]FDG) and fibroblast activation protein inhibitor ([⁶⁸Ga]Ga-FAPI-04), discordant lesions (FDG+/FAPI-) in aggressive neuroendocrine neoplasms (NENs) are linked to shorter progression-free survival (PFS). This study evaluated the prognostic value of such lesions in comparison to histological fibroblast activation protein (FAP) expression from a clinically obtained biopsy and their impact on PFS. Methods: 23 patients with aggressive NENs underwent both [¹⁸F]FDG and [⁶⁸Ga]Ga-FAPI-04 PET/CT as well as biopsy within a short period of time. PET parameters [standardized uptake values (SUV): SUVmax, SUVmean, SUVpeak] were measured, as were tumor volume (TV), and total lesion uptake (TLU = TV × SUVmean). FDG+/FAPI- lesions were identified. FAP expression was assessed immunohistochemically using the immunoreactive score (IRS-FAP). Correlations between PET metrics, IRS-FAP, and FDG+/FAPI- lesions were analyzed. Cox regression and log-rank test were used to evaluate associations with PFS. Results: IRS-FAP correlated significantly with TV ([¹⁸F]FDG: p=0.0165; [⁶⁸Ga]Ga-FAPI-04: p = 0.0181) and TLU ([¹⁸F]FDG: p = 0.0170; [⁶⁸Ga]Ga-FAPI-04: p = 0.0253). There was no significant correlation for IRS-FAP with SUV parameters. FDG+/FAPI- lesions were found in 9/23 patients and associated with significantly shorter PFS (4 vs. 10 months, HR: 3.383, p = 0.0015). [¹⁸F]FDG-TV and [¹⁸F]FDG-TLU correlated with PFS, but IRS-FAP showed no significant association with either PFS or FDG+/FAPI- lesions. Conclusions: FAP expression on immunohistochemistry obtained from a single biopsy site does not predict discordant PET/CT findings. In contrast, the presence of FDG+/FAPI- lesions is a strong prognostic factor for reduced PFS. Thus, dual-tracer PET/CT may offer superior risk stratification compared to single-lesion histological FAP assessment in aggressive NENs.