Ischaemic stroke, a leading cause of death and disability, critically relies on neuroimaging for characterising the anatomical pattern of injury. Diffusion-weighted imaging (DWI) provides the highest expressivity in ischemic stroke but poses substantial challenges for automated lesion segmentation: susceptibility artefacts, morphological heterogeneity, age-related comorbidities, time-dependent signal dynamics, instrumental variability, and limited labelled data. Current U-Net-based models therefore underperform, a problem accentuated by inadequate evaluation metrics that focus on mean performance, neglecting anatomical, subpopulation, and acquisition-dependent variability. Here, we present a high-performance DWI lesion segmentation tool addressing these challenges through optimized vision transformer-based architectures, integration of 3563 annotated lesions from multi-site data, and algorithmic enhancements, achieving state-of-the-art results. We further propose a novel evaluative framework assessing model fidelity, equity (across demographics and lesion subtypes), anatomical precision, and robustness to instrumental variability, promoting clinical and research utility. This work advances stroke imaging by reconciling model expressivity with domain-specific challenges and redefining performance benchmarks to prioritize equity and generalizability, critical for personalized medicine and mechanistic research.

The integration of AI into radiology introduces opportunities for improved clinical care provision and efficiency but it demands a meticulous approach to mitigate potential risks as with any other new technology. Beginning with rigorous pre-deployment evaluation and validation, the focus should be on ensuring models meet the highest standards of safety, effectiveness and efficacy for their intended applications. Input and output guardrails implemented during production usage act as an additional layer of protection, identifying and addressing individual failures as they occur. Continuous post-deployment monitoring allows for tracking population-level performance (data drift), fairness, and value delivery over time. Scheduling reviews of post-deployment model performance and educating radiologists about new algorithmic-driven findings is critical for AI to be effective in clinical practice. Recognizing that no single AI solution can provide absolute assurance even when limited to its intended use, the synergistic application of quality assurance at multiple levels - regulatory, clinical, technical, and ethical - is emphasized. Collaborative efforts between stakeholders spanning healthcare systems, industry, academia, and government are imperative to address the multifaceted challenges involved. Trust in AI is an earned privilege, contingent on a broad set of goals, among them transparently demonstrating that the AI adheres to the same rigorous safety, effectiveness and efficacy standards as other established medical technologies. By doing so, developers can instil confidence among providers and patients alike, enabling the responsible scaling of AI and the realization of its potential benefits. The roadmap presented herein aims to expedite the achievement of deployable, reliable, and safe AI in radiology.

The architecture of the brain is too complex to be intuitively surveyable without the use of compressed representations that project its variation into a compact, navigable space. The task is especially challenging with high-dimensional data, such as gene expression, where the joint complexity of anatomical and transcriptional patterns demands maximum compression. Established practice is to use standard principal component analysis (PCA), whose computational felicity is offset by limited expressivity, especially at great compression ratios. Employing whole-brain, voxel-wise Allen Brain Atlas transcription data, here we systematically compare compressed representations based on the most widely supported linear and non-linear methods-PCA, kernel PCA, non-negative matrix factorization (NMF), t-stochastic neighbour embedding (t-SNE), uniform manifold approximation and projection (UMAP), and deep auto-encoding-quantifying reconstruction fidelity, anatomical coherence, and predictive utility with respect to signalling, microstructural, and metabolic targets. We show that deep auto-encoders yield superior representations across all metrics of performance and target domains, supporting their use as the reference standard for representing transcription patterns in the human brain.

The quantification of cognitive powers rests on identifying a behavioural task that depends on them. Such dependence cannot be assured, for the powers a task invokes cannot be experimentally controlled or constrained a priori, resulting in unknown vulnerability to failure of specificity and generalisability. Evaluating a compact version of Raven's Advanced Progressive Matrices (RAPM), a widely used clinical test of fluid intelligence, we show that LaMa, a self-supervised artificial neural network trained solely on the completion of partially masked images of natural environmental scenes, achieves human-level test scores a prima vista, without any task-specific inductive bias or training. Compared with cohorts of healthy and focally lesioned participants, LaMa exhibits human-like variation with item difficulty, and produces errors characteristic of right frontal lobe damage under degradation of its ability to integrate global spatial patterns. LaMa's narrow training and limited capacity -- comparable to the nervous system of the fruit fly -- suggest RAPM may be open to computationally simple solutions that need not necessarily invoke abstract reasoning.

We propose a new method, Patch-CNN, for diffusion tensor (DT) estimation from only six-direction diffusion weighted images (DWI). Deep learning-based methods have been recently proposed for dMRI parameter estimation, using either voxel-wise fully-connected neural networks (FCN) or image-wise convolutional neural networks (CNN). In the acute clinical context -- where pressure of time limits the number of imaged directions to a minimum -- existing approaches either require an infeasible number of training images volumes (image-wise CNNs), or do not estimate the fibre orientations (voxel-wise FCNs) required for tractogram estimation. To overcome these limitations, we propose Patch-CNN, a neural network with a minimal (non-voxel-wise) convolutional kernel (3$\times$3$\times$3). Compared with voxel-wise FCNs, this has the advantage of allowing the network to leverage local anatomical information. Compared with image-wise CNNs, the minimal kernel vastly reduces training data demand. Evaluated against both conventional model fitting and a voxel-wise FCN, Patch-CNN, trained with a single subject is shown to improve the estimation of both scalar dMRI parameters and fibre orientation from six-direction DWIs. The improved fibre orientation estimation is shown to produce improved tractogram.

Causal mapping of the functional organisation of the human brain requires evidence of \textit{necessity} available at adequate scale only from pathological lesions of natural origin. This demands inferential models with sufficient flexibility to capture both the observable distribution of pathological damage and the unobserved distribution of the neural substrate. Current model frameworks -- both mass-univariate and multivariate -- either ignore distributed lesion-deficit relations or do not model them explicitly, relying on featurization incidental to a predictive task. Here we initiate the application of deep generative neural network architectures to the task of lesion-deficit inference, formulating it as the estimation of an expressive hierarchical model of the joint lesion and deficit distributions conditioned on a latent neural substrate. We implement such deep lesion deficit inference with variational convolutional volumetric auto-encoders. We introduce a comprehensive framework for lesion-deficit model comparison, incorporating diverse candidate substrates, forms of substrate interactions, sample sizes, noise corruption, and population heterogeneity. Drawing on 5500 volume images of ischaemic stroke, we show that our model outperforms established methods by a substantial margin across all simulation scenarios, including comparatively small-scale and noisy data regimes. Our analysis justifies the widespread adoption of this approach, for which we provide an open source implementation: https://github.com/guilherme-pombo/vae_lesion_deficit

In a clinical setting it is essential that deployed image processing systems are robust to the full range of inputs they might encounter and, in particular, do not make confidently wrong predictions. The most popular approach to safe processing is to train networks that can provide a measure of their uncertainty, but these tend to fail for inputs that are far outside the training data distribution. Recently, generative modelling approaches have been proposed as an alternative; these can quantify the likelihood of a data sample explicitly, filtering out any out-of-distribution (OOD) samples before further processing is performed. In this work, we focus on image segmentation and evaluate several approaches to network uncertainty in the far-OOD and near-OOD cases for the task of segmenting haemorrhages in head CTs. We find all of these approaches are unsuitable for safe segmentation as they provide confidently wrong predictions when operating OOD. We propose performing full 3D OOD detection using a VQ-GAN to provide a compressed latent representation of the image and a transformer to estimate the data likelihood. Our approach successfully identifies images in both the far- and near-OOD cases. We find a strong relationship between image likelihood and the quality of a model's segmentation, making this approach viable for filtering images unsuitable for segmentation. To our knowledge, this is the first time transformers have been applied to perform OOD detection on 3D image data. Code is available at github.com/marksgraham/transformer-ood.

Out-of-distribution detection is crucial to the safe deployment of machine learning systems. Currently, unsupervised out-of-distribution detection is dominated by generative-based approaches that make use of estimates of the likelihood or other measurements from a generative model. Reconstruction-based methods offer an alternative approach, in which a measure of reconstruction error is used to determine if a sample is out-of-distribution. However, reconstruction-based approaches are less favoured, as they require careful tuning of the model's information bottleneck - such as the size of the latent dimension - to produce good results. In this work, we exploit the view of denoising diffusion probabilistic models (DDPM) as denoising autoencoders where the bottleneck is controlled externally, by means of the amount of noise applied. We propose to use DDPMs to reconstruct an input that has been noised to a range of noise levels, and use the resulting multi-dimensional reconstruction error to classify out-of-distribution inputs. We validate our approach both on standard computer-vision datasets and on higher dimension medical datasets. Our approach outperforms not only reconstruction-based methods, but also state-of-the-art generative-based approaches. Code is available at https://github.com/marksgraham/ddpm-ood.

The complex heterogeneity of brain tumours is increasingly recognized to demand data of magnitudes and richness only fully-inclusive, large-scale collections drawn from routine clinical care could plausibly offer. This is a task contemporary machine learning could facilitate, especially in neuroimaging, but its ability to deal with incomplete data common in real world clinical practice remains unknown. Here we apply state-of-the-art methods to large scale, multi-site MRI data to quantify the comparative fidelity of automated tumour segmentation models replicating the various levels of sequence availability observed in the clinical reality. We compare deep learning (nnU-Net-derived) segmentation models with all possible combinations of T1, contrast-enhanced T1, T2, and FLAIR sequences, trained and validated with five-fold cross-validation on the 2021 BraTS-RSNA glioma population of 1251 patients, with further testing on a real-world 50 patient sample diverse in not only MRI scanner and field strength, but a random selection of pre- and post-operative imaging also. Models trained on incomplete imaging data segmented lesions well, often equivalently to those trained on complete data, exhibiting Dice coefficients of 0.907 (single sequence) to 0.945 (full datasets) for whole tumours, and 0.701 (single sequence) to 0.891 (full datasets) for component tissue types. Incomplete data segmentation models could accurately detect enhancing tumour in the absence of contrast imaging, quantifying its volume with an R2 between 0.95-0.97, and were invariant to lesion morphometry. Deep learning segmentation models characterize tumours well when missing data and can even detect enhancing tissue without the use of contrast. This suggests translation to clinical practice, where incomplete data is common, may be easier than hitherto believed, and may be of value in reducing dependence on contrast use.

Models of human motion commonly focus either on trajectory prediction or action classification but rarely both. The marked heterogeneity and intricate compositionality of human motion render each task vulnerable to the data degradation and distributional shift common to real-world scenarios. A sufficiently expressive generative model of action could in theory enable data conditioning and distributional resilience within a unified framework applicable to both tasks. Here we propose a novel architecture based on hierarchical variational autoencoders and deep graph convolutional neural networks for generating a holistic model of action over multiple time-scales. We show this Hierarchical Graph-convolutional Variational Autoencoder (HG-VAE) to be capable of generating coherent actions, detecting out-of-distribution data, and imputing missing data by gradient ascent on the model's posterior. Trained and evaluated on H3.6M and the largest collection of open source human motion data, AMASS, we show HG-VAE can facilitate downstream discriminative learning better than baseline models.