Purpose: A magnetic resonance imaging (MRI) exam typically consists of several sequences that yield different image contrasts. Each sequence is parameterized through multiple acquisition parameters that influence image contrast, signal-to-noise ratio, acquisition time, and/or resolution. Depending on the clinical indication, different contrasts are required by the radiologist to make a diagnosis. As MR sequence acquisition is time consuming and acquired images may be corrupted due to motion, a method to synthesize MR images with adjustable contrast properties is required. Methods: Therefore, we trained an image-to-image generative adversarial network conditioned on the MR acquisition parameters repetition time and echo time. Our approach is motivated by style transfer networks, whereas the “style” for an image is explicitly given in our case, as it is determined by the MR acquisition parameters our network is conditioned on. Results: This enables us to synthesize MR images with adjustable image contrast. We evaluated our approach on the fastMRI dataset, a large set of publicly available MR knee images, and show that our method outperforms a benchmark pix2pix approach in the translation of non-fat-saturated MR images to fat-saturated images. Our approach yields a peak signal-to-noise ratio and structural similarity of 24.48 and 0.66, surpassing the pix2pix benchmark model significantly. Conclusion: Our model is the first that enables fine-tuned contrast synthesis, which can be used to synthesize missing MR-contrasts or as a data augmentation technique for AI training in MRI. It can also be used as basis for other image-to-image translation tasks within medical imaging, e.g., to enhance intermodality translation (MRI → CT) or 7 T image synthesis from 3 T MR images.

Automated protocoling for MRI examinations is an amendable target for workflow automation with artificial intelligence. However, there are still challenges to overcome for a successful and robust approach. These challenges are outlined and analyzed in this work. Through a literature review, we analyzed limitations of currently published approaches for automated protocoling. Then, we assessed these limitations quantitatively based on data from a private radiology practice. For this, we assessed the information content provided by the clinical indication by computing the overlap coefficients for the sets of ICD-10-coded admitting diagnoses of different MRI protocols. Additionally, we assessed the heterogeneity of protocol trees from three different MRI scanners based on the overlap coefficient, on MRI protocol and sequence level. Additionally, we applied sequence name standardization to demonstrate its effect on the heterogeneity assessment, i.e., the overlap coefficient, of different protocol trees. The overlap coefficient for the set of ICD-10-coded admitting diagnoses for different protocols ranges from 0.14 to 0.56 for brain/head MRI exams and 0.04 to 0.57 for spine exams. The overlap coefficient across the set of sequences used at two different scanners increases when applying sequence name standardization (from 0.81/0.86 to 0.93). Automated protocoling for MRI examinations has the potential to reduce the workload for radiologists. However, an automated protocoling approach cannot be solely based on admitting diagnosis as it does not provide sufficient information. Moreover, sequence name standardization increases the overlap coefficient across the set of sequences used at different scanners and therefore facilitates transfer learning.

Although the level of digitalization and automation steadily increases in radiology, billing coding for magnetic resonance imaging (MRI) exams in the radiology department is still based on manual input from the technologist. After the exam completion, the technologist enters the corresponding exam codes that are associated with billing codes in the radiology information system. Moreover, additional billing codes are added or removed, depending on the performed procedure. This workflow is time-consuming and we showed that billing codes reported by the technologists contain errors. The coding workflow can benefit from an automated system, and thus a prediction model for automated assignment of billing codes for MRI exams based on MRI log data is developed in this work. To the best of our knowledge, it is the first attempt to focus on the prediction of billing codes from modality log data. MRI log data provide a variety of information, including the set of executed MR sequences, MR scanner table movements, and given a contrast medium. MR sequence names are standardized using a heuristic approach and incorporated into the features for the prediction. The prediction model is trained on 9754 MRI exams and tested on 1 month of log data (423 MRI exams) from two MRI scanners of the radiology site for the Swiss medical tariffication system Tarmed. The developed model, an ensemble of classifier chains with multilayer perceptron as a base classifier, predicts medical billing codes for MRI exams with a micro-averaged F1-score of 97.8% (recall 98.1%, precision 97.5%). Manual coding reaches a micro-averaged F1-score of 98.1% (recall 97.4%, precision 98.8%). Thus, the performance of automated coding is close to human performance. Integrated into the clinical environment, this work has the potential to free the technologist from a non-value adding an administrative task, therefore enhance the MRI workflow, and prevent coding errors.

A magnetic resonance imaging (MRI) exam typically consists of the acquisition of multiple MR pulse sequences, which are required for a reliable diagnosis. With the rise of generative deep learning models, approaches for the synthesis of MR images are developed to either synthesize additional MR contrasts, generate synthetic data, or augment existing data for AI training. While current generative approaches allow only the synthesis of specific sets of MR contrasts, we developed a method to generate synthetic MR images with adjustable image contrast. Therefore, we trained a generative adversarial network (GAN) with a separate auxiliary classifier (AC) network to generate synthetic MR knee images conditioned on various acquisition parameters (repetition time, echo time, and image orientation). The AC determined the repetition time with a mean absolute error (MAE) of 239.6 ms, the echo time with an MAE of 1.6 ms, and the image orientation with an accuracy of 100%. Therefore, it can properly condition the generator network during training. Moreover, in a visual Turing test, two experts mislabeled 40.5% of real and synthetic MR images, demonstrating that the image quality of the generated synthetic and real MR images is comparable. This work can support radiologists and technologists during the parameterization of MR sequences by previewing the yielded MR contrast, can serve as a valuable tool for radiology training, and can be used for customized data generation to support AI training.

Accurately labeling outcomes in real‐world data for machine learning is challenging due to data sparsity and imbalances. This study developed and evaluated a pharmacokinetic–pharmacodynamic (PKPD)‐informed labeling strategy to enhance the risk prediction of docetaxel‐induced neutropenia. Machine learning models were trained on real‐world data from 4,248 patients using two approaches for comparison. The “naive” labeling method used only neutrophil observations, while the “PKPD‐informed” method used simulations from a semi‐mechanistic model to determine the neutrophil nadir for each treatment cycle. Three machine learning models (logistic regression, XGBoost, TabPFN) were trained with baseline laboratory data to predict severe neutropenia (neutrophil count <0.1 cells × 10 ⁹ /L) prior to the first docetaxel dose. The PKPD labeling approach enabled the labeling of 3.4 times more patient instances (7,719 vs. 2,283) than the naive method. Across all machine learning architectures, models trained with PKPD‐informed labels demonstrated significantly superior predictive performance (AUC‐ROC and AUC‐PR) compared to those trained with naive labels. This advantage was maintained even when training set sizes were matched. PKPD‐informed labeling overcomes limitations of sparse real‐world data, increasing both the quantity and apparent quality of labels for machine learning model training. This methodology enhances the performance of machine learning models for predicting severe neutropenia and represents a robust, generalizable framework for improving clinical outcome prediction.