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.
BONSAI is an end-to-end Python framework for processing and analysing Electronic Health Records (EHR) data. It extends the model described in the CORE-BEHRT paper (Odgaard et al., 2024) and is designed to streamline data preparation, model pre-training, and fine-tuning for patient-level prediction tasks. The framework is built for efficient large-scale processing of EHR data, making it suitable for clinical applications involving substantial volumes of patient records. This enables models for patient outcome prediction and supports scalable clinical research.
Large-scale mega-analyses of worldwide combined Magnetic Resonance Imaging (MRI) studies have demonstrated brain differences between individuals with mental disorders and controls. However, the potential of large-scale observational studies using population-based clinical MRI data remains unexplored. We analyzed clinical MRI data from 23,545 patients in the Eastern half of Denmark (Capital Region of Denmark and Region Zealand). 2,774 patients with mental disorders and 2,062 non-psychiatric controls fulfilled our inclusion and exclusion criteria. Patients with mental disorders exhibited smaller thalamic (d=−0.298) and amygdala volumes (d=−0.250), with larger ventricles (d=0.272), and thinner insula (d=−0.177), all p<0.0001. Analysis across all ROIs revealed a widespread pattern of thinner cortex (d=−0.180), especially in the temporal pole (d=−0.234) and superior frontal (d=−0.212) regions, and increased extracerebral cerebrospinal fluid (d=0.264). For volumetric measurements, findings were consistent across different inclusion and exclusion criteria but varied for cortical thickness measurements. Utilizing this currently largest population-based MRI cohort for mental disorders, we demonstrate that clinical MRI scans can detect brain structural differences among patients with mental disorders in real-world clinical settings, aiding in the stratification of patients without mental disorders. Cross-disorder analyses reveal shared neuroanatomical changes, including globally smaller brain volumes and thinner cortex. Integrating large-scale clinical MRI data with electronic health records holds promise for improved patient stratification and tracking of disease progression for future longitudinal cross-disorder studies, bridging real-world MRI data with clinical trajectories for further biological subgrouping.