Quality control of MRI is essential for excluding problematic acquisitions and avoiding bias in subsequent image processing and analysis. Visual inspection is subjective and impractical for large scale datasets. Although automated quality assessments have been demonstrated on single-site datasets, it is unclear that solutions can generalize to unseen data acquired at new sites. Here, we introduce the MRI Quality Control tool (MRIQC), a tool for extracting quality measures and fitting a binary (accept/exclude) classifier. Our tool can be run both locally and as a free online service via the OpenNeuro.org portal. The classifier is trained on a publicly available, multi-site dataset (17 sites, N = 1102). We perform model selection evaluating different normalization and feature exclusion approaches aimed at maximizing across-site generalization and estimate an accuracy of 76%±13% on new sites, using leave-one-site-out cross-validation. We confirm that result on a held-out dataset (2 sites, N = 265) also obtaining a 76% accuracy. Even though the performance of the trained classifier is statistically above chance, we show that it is susceptible to site effects and unable to account for artifacts specific to new sites. MRIQC performs with high accuracy in intra-site prediction, but performance on unseen sites leaves space for improvement which might require more labeled data and new approaches to the between-site variability. Overcoming these limitations is crucial for a more objective quality assessment of neuroimaging data, and to enable the analysis of extremely large and multi-site samples.

A central goal of cognitive neuroscience is to decode human brain activity—that is, to infer mental processes from observed patterns of whole-brain activation. Previous decoding efforts have focused on classifying brain activity into a small set of discrete cognitive states. To attain maximal utility, a decoding framework must be open-ended, systematic, and context-sensitive—that is, capable of interpreting numerous brain states, presented in arbitrary combinations, in light of prior information. Here we take steps towards this objective by introducing a probabilistic decoding framework based on a novel topic model—Generalized Correspondence Latent Dirichlet Allocation—that learns latent topics from a database of over 11,000 published fMRI studies. The model produces highly interpretable, spatially-circumscribed topics that enable flexible decoding of whole-brain images. Importantly, the Bayesian nature of the model allows one to “seed” decoder priors with arbitrary images and text—enabling researchers, for the first time, to generate quantitative, context-sensitive interpretations of whole-brain patterns of brain activity. A central goal of cognitive neuroscience is to decode human brain activity—i.e., to be able to infer mental processes from observed patterns of whole-brain activity. However, existing approaches to brain decoding suffer from a number of important limitations—for example, they often work only in one narrow domain of cognition, and cannot be easily generalized to novel contexts. Here we address such limitations by introducing a simple probabilistic framework based on a novel topic modeling approach. We use our approach to extract a set of highly interpretable latent “topics” from a large meta-analytic database of over 11,000 published fMRI studies. Each topic is associated with a single brain region and a set of semantically coherent cognitive functions. We demonstrate how these topics can be used to automatically “decode” brain activity in an open-ended way, enabling researchers to draw tentative conclusions about mental function on the basis of virtually any pattern of whole-brain activity. We highlight several important features of our framework, notably including the ability to take into account knowledge of the experimental context and/or prior experimenter belief.

Artificial intelligence and machine learning (AI/ML) in medical imaging provide important methods for leveraging large amounts of data to build models to detect disease and provide diagnosis, prognosis, and risk assessment tools to support decision-making in medicine. The development and deployment of AI/ML models in a wide variety of fields have grown substantially in the past decade, broadening considerations for more careful evaluation of the algorithms. Researchers and developers note that there are persistent challenges in generalizability and bias along the model development and deployment pipeline, including the characteristics of data used in training models and testing them. These issues have been noted and studied in fields, such as economics and sociology¹ as well as the natural sciences.² Efforts to understand and address bias in the broader AI/ML literature involve expanding fairness metrics and definitions to include additional factors³ and the development of mitigation strategies. These concerns are especially important in healthcare and medicine broadly⁴ and have been an area of study in medical imaging by several groups and initiatives, including studies which demonstrate the potential for bias in medical imaging of COVID-19.⁵ Recently, a three-part series^{6 – 8} published in Radiology: Artificial Intelligence presented methods and insights for mitigating bias in AI/ML for medical imaging through considerations for data handling,⁶ model development,⁷ and performance metrics.⁸

Deep learning (DL) models can harness electronic health records (EHRs) to predict diseases and extract radiologic findings for diagnosis. With ambulatory chest radiographs (CXRs) frequently ordered, we investigated detecting type 2 diabetes (T2D) by combining radiographic and EHR data using a DL model. Our model, developed from 271,065 CXRs and 160,244 patients, was tested on a prospective dataset of 9,943 CXRs. Here we show the model effectively detected T2D with a ROC AUC of 0.84 and a 16% prevalence. The algorithm flagged 1,381 cases (14%) as suspicious for T2D. External validation at a distinct institution yielded a ROC AUC of 0.77, with 5% of patients subsequently diagnosed with T2D. Explainable AI techniques revealed correlations between specific adiposity measures and high predictivity, suggesting CXRs’ potential for enhanced T2D screening.