DeepCEST 7 T: Fast and homogeneous mapping of 7 T CEST MRI parameters and their uncertainty quantification.

PURPOSE: In this work, we investigated the ability of neural networks to rapidly and robustly predict Lorentzian parameters of multi-pool CEST MRI spectra at 7 T with corresponding uncertainty maps to make them quickly and easily available for routine clinical use. METHODS: We developed a deepCEST 7 T approach that generates CEST contrasts from just 1 scan with robustness against B1 inhomogeneities. The input data for a neural feed-forward network consisted of 7 T in vivo uncorrected Z-spectra of a single B1 level, and a B1 map. The 7 T raw data were acquired using a 3D snapshot gradient echo multiple interleaved mode saturation CEST sequence. These inputs were mapped voxel-wise to target data consisting of Lorentzian amplitudes generated conventionally by 5-pool Lorentzian fitting of normalized, denoised, B0 - and B1 -corrected Z-spectra. The deepCEST network was trained with Gaussian negative log-likelihood loss, providing an uncertainty quantification in addition to the Lorentzian amplitudes. RESULTS: The deepCEST 7 T network provides fast and accurate prediction of all Lorentzian parameters also when only a single B1 level is used. The prediction was highly accurate with respect to the Lorentzian fit amplitudes, and both healthy tissues and hyperintensities in tumor areas are predicted with a low uncertainty. In corrupted cases, high uncertainty indicated wrong predictions reliably. CONCLUSION: The proposed deepCEST 7 T approach reduces scan time by 50% to now 6:42 min, but still delivers both B0 - and B1 -corrected homogeneous CEST contrasts along with an uncertainty map, which can increase diagnostic confidence. Multiple accurate 7 T CEST contrasts are delivered within seconds.

COVID-19-associated cerebral microbleeds in the general population.

Cerebral microbleeds are frequent incidental findings on brain MRI and have previously been shown to occur in Coronavirus Disease 2019 (COVID-19) cohorts of critically ill patients. We aimed to determine the risk of having microbleeds on medically indicated brain MRI and compare non-hospitalized COVID-19-infected patients with non-infected controls. In this retrospective case-control study, we included patients over 18 years of age, having an MRI with a susceptibility-weighted sequence, between 1 January 2019 and 1 July 2021. Cases were identified based on a positive reverse transcriptase polymerase chain reaction test for SARS-CoV-2 and matched with three non-exposed controls, based on age, sex, body mass index and comorbidities. The number of cerebral microbleeds on each scan was determined using artificial intelligence. We included 73 cases and 219 matched non-exposed controls. COVID-19 was associated with significantly greater odds of having cerebral microbleeds on MRI [odds ratio 2.66 (1.23-5.76, 95% confidence interval)], increasingly so when patients with dementia and hospitalized patients were excluded. Our findings indicate that cerebral microbleeds may be associated with COVID-19 infections. This finding may add to the pathophysiological considerations of cerebral microbleeds and help explain cases of incidental cerebral microbleeds in patients with previous COVID-19.

Assessing optimal methods for transferring machine learning models to low-volume and imbalanced clinical datasets: experiences from predicting outcomes of Danish trauma patients.

Introduction: Accurately predicting patient outcomes is crucial for improving healthcare delivery, but large-scale risk prediction models are often developed and tested on specific datasets where clinical parameters and outcomes may not fully reflect local clinical settings. Where this is the case, whether to opt for de-novo training of prediction models on local datasets, direct porting of externally trained models, or a transfer learning approach is not well studied, and constitutes the focus of this study. Using the clinical challenge of predicting mortality and hospital length of stay on a Danish trauma dataset, we hypothesized that a transfer learning approach of models trained on large external datasets would provide optimal prediction results compared to de-novo training on sparse but local datasets or directly porting externally trained models. Methods: Using an external dataset of trauma patients from the US Trauma Quality Improvement Program (TQIP) and a local dataset aggregated from the Danish Trauma Database (DTD) enriched with Electronic Health Record data, we tested a range of model-level approaches focused on predicting trauma mortality and hospital length of stay on DTD data. Modeling approaches included de-novo training of models on DTD data, direct porting of models trained on TQIP data to the DTD, and a transfer learning approach by training a model on TQIP data with subsequent transfer and retraining on DTD data. Furthermore, data-level approaches, including mixed dataset training and methods countering imbalanced outcomes (e.g., low mortality rates), were also tested. Results: Using a neural network trained on a mixed dataset consisting of a subset of TQIP and DTD, with class weighting and transfer learning (retraining on DTD), we achieved excellent results in predicting mortality, with a ROC-AUC of 0.988 and an F2-score of 0.866. The best-performing models for predicting long-term hospitalization were trained only on local data, achieving an ROC-AUC of 0.890 and an F1-score of 0.897, although only marginally better than alternative approaches. Conclusion: Our results suggest that when assessing the optimal modeling approach, it is important to have domain knowledge of how incidence rates and workflows compare between hospital systems and datasets where models are trained. Including data from other health-care systems is particularly beneficial when outcomes are suffering from class imbalance and low incidence. Scenarios where outcomes are not directly comparable are best addressed through either de-novo local training or a transfer learning approach.