Steve Pieper

Purpose: Clinical assessment of vertebral lesion quality (osteolytic, osteoblastic, mixed) remains subjective, with limited interobserver reliability. This study evaluated a novel application of 3D convolutional neural networks (3D-CNNs) for classifying lesion quality from CT volumes in metastatic cancer patients. Materials and Methods: This retrospective study used CT data from 151 cancer patients planned for radiotherapy for metastatic spine disease (September 2020-July 2024). Leveraging vertebra-level expert annotations, we introduced an unconventional U-Net-based strategy converting coarse voxel-wise predictions into vertebra-level lesion classifications. The final dataset comprised 2,125 vertebrae across four classes (no lesion, osteolytic, osteoblastic, mixed), split into a 3-fold cross-validation set and an independent holdout test set. Model performance was benchmarked against a DenseNet121 baseline and a musculoskeletal radiologist, with Cohen's kappa assessing inter-rater agreement. Results: The 3D model achieved an ensemble accuracy of 84.7%, outperforming DenseNet121 (72.1%), with substantial gains in F1 score, precision, and balanced accuracy. It showed high concordance with the radiologist (Cohen's kappa = 0.76) and comparable sensitivity and specificity across all lesion subtypes. We found both models and the radiologist to struggle with osteolytic lesions, reflecting the difficulty of distinguishing this class from age-related changes in vertebral bone density and architecture caused by benign bone lesions, age-related systemic skeletal disorders and cancer treatments. Conclusions: 3D-CNNs trained with vertebra-level labels can accurately and reliably classify vertebral metastatic lesion quality from CT scans, offering a scalable path toward automated characterization of metastatic spine disease to support clinical decision-making and large-scale radiomics research.

Artificial intelligence (AI) has the potential to transform medical imaging by automating image analysis and accelerating clinical research. However, research and clinical use are limited by the wide variety of AI implementations and architectures, inconsistent documentation, and reproducibility issues. Here, we introduce MHub.ai, an open-source, container-based platform that standardizes access to AI models with minimal configuration, promoting accessibility and reproducibility in medical imaging. MHub.ai packages models from peer-reviewed publications into standardized containers that support direct processing of DICOM and other formats, provide a unified application interface, and embed structured metadata. Each model is accompanied by publicly available reference data to confirm model operation. MHub.ai includes an initial set of segmentation, prediction, and feature extraction models for different modalities. The modular framework enables adaptation of any model and supports community contributions. We demonstrate the utility of the platform through comparative evaluation of lung segmentation models on public clinical data, and publicly release the generated segmentations and evaluation metrics as interactive dashboards to emphasize transparency and to enable case-level inspection. By enabling side-by-side benchmarking with identical execution commands and standardized outputs, MHub.ai reduces technical friction for model execution, evaluation, and comparison.

The increasing use of 3D imaging technologies in biological sciences is generating vast repositories of anatomical data, yet significant barriers prevent these data from reaching its full potential in educational and collaborative contexts. While sharing raw computed tomography and magnetic resonance imaging scans has become routine, distributing value‐added segmented datasets—where anatomical structures are precisely labelled and delineated—remains difficult and rare. Current repositories function primarily as static archives, lacking mechanisms for iterative refinement, community‐driven curation and the controlled terminology essential for downstream computational applications, including artificial intelligence (AI). We propose that segmented 3D morphological datasets should be reconceptualized as collaborative, version‐controlled projects rather than static archival products. We introduce MorphoDepot, a framework that adapts the ‘fork‐and‐contribute’ model—a cornerstone of modern open‐source software development—for collaborative management of 3D morphological data. By integrating git version control and GitHub's collaborative infrastructure with 3D Slicer and its SlicerMorph extension, MorphoDepot transforms segmented anatomical datasets into dynamic, community‐curated projects. By treating each anatomical dataset as an individual repository that can be forked, modified, reviewed and merged, MorphoDepot enables communities of researchers, educators and students to co‐create high‐quality, standardized anatomical atlases through a transparent, auditable process. This approach directly addresses the challenges of distributed collaboration, enforces transparent provenance tracking and creates high‐quality, standardized training data for AI model development. The result is a system that embodies FAIR (Findable, Accessible, Interoperable and Reusable) data principles while creating powerful new opportunities for remote learning and collaborative science. We argue that format accessibility and collaborative curation are inseparable, genuine community‐driven data refinement requires open formats and open tools as prerequisites for participation.

3D Slicer is an open-source software platform for the analysis, segmentation, and three-dimensional visualization of medical imaging data. Although the platform is used by an international research community, its interface was historically available primarily in English, which may limit accessibility for non-English-speaking users. This study describes the development of an ad hoc methodology for the Brazilian Portuguese localization of 3D Slicer within the broader Latin American localization initiative. The methodology addresses recurrent linguistic challenges identified in a preliminary corpus of 300 interface strings, including domain-specific vocabulary, acronyms, word order, passive voice, syntagms, and the adaptation of technical terms. The translation process emphasizes textual uniformity, cohesion, terminological accuracy, and contextual validation in biomedical-computational environments. The proposed framework may support similar software localization efforts in other non-English-speaking contexts, especially when technical precision and linguistic adaptation must be balanced.

While deeper white matter connections, such as the arcuate fasciculus and frontal aslant tract, are well known for their role in language and show leftward asymmetries in adults, the contribution of the short-range cortico-cortical superficial white matter (SWM) connections remains less understood. In this preregistered study, we examined white matter connections of Broca’s area and its right hemisphere homolog in early adolescents and young adults using two large, publicly available datasets: the Adolescent Brain Cognitive Development Study and the Human Connectome Project Young Adult Study, totaling over 10,000 participants. We anatomically curated the O’Donnell Research Group (ORG) tractography atlas to identify SWM fiber clusters intersecting Broca’s area (pars opercularis and pars triangularis), confirmed through expert visual inspection. We investigated the microstructure, structural connectivity, and lateralization of Broca’s area SWM and its relationship with language performance (Picture Vocabulary and Oral Reading Recognition assessments), in comparison to the deeper white matter connections of the frontal aslant tract and arcuate fasciculus. The arcuate fasciculus demonstrated the strongest and most consistent leftward lateralization of both microstructure (fractional anisotropy, FA) and structural connectivity (number of streamlines, NoS), with structure-function associations that were bilateral in adolescents and left-dominant in adults. Interestingly, despite weaker lateralization than the arcuate fasciculus, both the SWM and the frontal aslant tract demonstrated comparable associations with language performance. The frontal aslant tract showed greater leftward lateralization (FA and NoS) with age and was bilaterally associated with language performance, particularly in adolescents. Compared to these deeper tracts, Broca’s area SWM demonstrated left-lateralized FA and right-lateralized NoS in both adolescents and adults, with stronger FA lateralization in adults. Bilateral SWM FA and NoS were significantly associated with language performance in both hemispheres and age groups. Overall, these results suggest that Broca’s area SWM may support language in a more bilaterally distributed manner and highlight the importance of considering SWM connections in studies of language development and neurosurgical planning.

Recent advances in deep learning have led to robust automated tools for segmentation of abdominal computed tomography (CT). Meanwhile, segmentation of magnetic resonance imaging (MRI) is substantially more challenging due to the inherent signal variability and the increased effort required for annotating training datasets. Hence, existing approaches are trained on limited sets of MRI sequences, which might limit their generalizability. To characterize the landscape of MRI abdominal segmentation tools, we present here a comprehensive benchmarking of the three state-of-the-art and open-source models: MRSegmentator, MRISegmentator-Abdomen, and TotalSegmentator MRI. Since these models are trained using labor-intensive manual annotation cycles, we also introduce and evaluate ABDSynth, a SynthSeg-based model purely trained on widely available CT segmentations (no real images). More generally, we assess accuracy and generalizability by leveraging three public datasets (not seen by any of the evaluated methods during their training), which span all major manufacturers, five MRI sequences, as well as a variety of subject conditions, voxel resolutions, and fields-of-view. Our results reveal that MRSegmentator achieves the best performance and is most generalizable. In contrast, ABDSynth yields slightly less accurate results, but its relaxed requirements in training data make it an alternative when the annotation budget is limited. The evaluation code and datasets are given for future benchmarking at https://github.com/deepakri201/AbdoBench, along with inference code and weights for ABDSynth.

The cerebellum, long implicated in movement, is now recognized as a contributor to higher-order cognition. The cerebellar pathways provide key structural links between the cerebellum and cerebral regions integral to language, memory, and executive function. Here, we present a large-scale, cross-sectional diffusion MRI (dMRI) analysis investigating the relationships between cerebellar pathway microstructure and cognitive performance in over 9,000 participants spanning pre-adolescence (n>8,000 from the ABCD dataset) and young adulthood (n>900 from the HCP-YA dataset). We assessed the microstructure of five cerebellar pathways—the inferior, middle, and superior cerebellar peduncles; the parallel fibers; and input/Purkinje fibers—using three dMRI measures of fractional anisotropy, mean diffusivity, and number of streamlines. Cognitive performance was evaluated using seven NIH Toolbox assessments of language, executive function, and memory. In both datasets, we found numerous significant associations between cerebellar pathway microstructure and cognitive performance. These associations showed a strong correlation across the two datasets (r = 0.47, p < 0.0001), underscoring the reliability of cerebellar dMRI-cognition relationships in pre-adolescents and young adults. In both datasets, the strongest associations were found between the superior cerebellar peduncle and performance on language assessments, suggesting this pathway plays an important role in language function across age groups. In young adults, but not pre-adolescents, parallel fiber microstructure was linked to inhibitory control, suggesting that contributions to attentional processes may emerge or strengthen with maturation. Overall, our findings highlight the important role of cerebellar pathways in cognition and the utility of large-scale datasets for advancing our understanding of brain-cognition relationships. Significance: This study provides strong evidence linking cerebellar pathway tissue microstructure to cognition across large populations of preadolescent children and young adults. By leveraging diffusion MRI tractography and cognitive performance data from two major datasets, we identify significant relationships between cerebellar pathway microstructure and cognitive performance. Importantly, findings are significantly correlated across datasets, pointing to the consistency of these relationships, bridging age groups and acquisitions. These results highlight the cerebellar pathways’ integral role in cognitive functioning and underscore the value of large-scale, population-based studies in advancing our understanding of brain-cognition relationships. Classification 1) Biological Sciences, Neuroscience; 2) Social Sciences, Psychological and Cognitive Sciences

Purpose Given the high prevalence of vertebral fractures post-radiotherapy in patients with metastatic spine disease, accurate and rapid muscle segmentation could support efforts to quantify muscular changes due to disease or treatment and enable biomechanical modeling for assessments of vertebral loading to improve personalized evaluation of vertebral fracture risk. This study presents a deep-learning approach for segmenting the complete volume of the trunk muscles from clinical CT images trained using sparsely annotated data. Materials and Methods we extracted 2,009 axial CT images at the midpoint of each vertebral level (T4 to L4) from clinical CT of 148 cancer patients. The key extensor and flexor muscles (up to 8 muscles per side) were manually contoured and labeled per image in the thoracic and lumbar regions. We first trained a 2D nnU-Net deep-learning model on these labels to segment key extensor and flexor muscles. Using these sparse annotations per spine, we trained the model to segment each muscle’s entire 3D volume Results The proposed method achieved comparable performance to manual segmentations, as assessed by expert radiologists, with a mean Dice score above 0.769. Significantly, the model drastically reduced segmentation time, from 4.3-6.5 hours for manual segmentation of 14 single axial CT images to approximately 1 minute for segmenting the complete thoracic-abdominal 3D volume. Conclusion The approach demonstrates high potential for automating 3D muscle segmentation, significantly reducing the manual intervention required for generating musculoskeletal models, and could be instrumental in enhancing clinical decision-making and patient care in radiation oncology. Summary A deep learning 2D nnU-Net model, trained on a sparse set of 2D muscle annotations, successfully segmented the entire volume of 20 thoracolumbar muscles from cancer patients’ clinical CT data. The model showed a remarkable increase in segmentation efficacy and generalizability, achieving comparable performance to manual segmentations in delineating each muscle anatomy. Key Points ▪ A deep learning model (2D nnU-Net), developed using a sparse set of single axial CT-slice at each mid-per vertebral level, containing manual image annotation of 20 thoracic and lumbar muscles, achieved comparable performance to manual segmentations, as assessed by expert radiologists, with a mean Dice score above 0.769. ▪ The model drastically reduced segmentation time, from 4.3-6.5 hours for manual segmentation of 14 single axial CT images to approximately 1 minute for segmenting the complete thoracic-abdominal 3D volume. ▪ Radiologist assessment based on a Likert scale (0-5) for clinical acceptability of the muscle anatomical segmentation showed strong model performance for a representative sample of clinical CT data (a (mean(SD) of 4.66 (0.73)) and external data (4.66 (0.73).