The researchers call their architecture ResNet50-MASAG-UNet, and each component of that name encodes a deliberate technical choice. The backbone is a UNet, the encoder-decoder network that has dominated biomedical segmentation since 2015, but with its standard encoder replaced by ResNet50, a 50-layer residual network pretrained on ImageNet. That transfer learning decision is crucial because the team evaluated their system on the DCA1 database, a public collection of just 134 grayscale angiograms of 300 by 300 pixels, each annotated with ground-truth vessel masks prepared by expert cardiologists at the Cardiology Department of the Mexican Social Security Institute. Small datasets like this cause deep encoders trained from scratch to overfit badly; the ablation numbers show it plainly. With random initialization the model reaches a Dice coefficient of 81.2 percent, but with ImageNet-pretrained weights it climbs to 88.3 percent—a gain of more than seven points attributable to pretraining alone.

Before any network sees the images, the pipeline applies a three-stage preprocessing sequence designed to rescue vessel visibility from the raw angiogram’s low signal-to-noise ratio. First, bilateral filtering with a neighbourhood diameter of 9 and color and spatial sigma values of 75 smooths background noise while preserving vessel edges, because bilateral filtering weights neighbours by both spatial distance and intensity similarity. Second, unsharp masking with a 9 by 9 Gaussian kernel, sigma of 10.0, and an amount factor of 1.5 sharpens fine vascular details by subtracting a blurred copy of the image from the original. Third, Contrast Limited Adaptive Histogram Equalization with a clip limit of 3.0 on an 8 by 8 tile grid redistributes local intensity contrast and corrects uneven brightness without amplifying noise in dark regions. The order matters: noise reduction comes first, edge sharpening second, local contrast enhancement last, so that CLAHE does not magnify artefacts the earlier steps have not yet suppressed. The preprocessing ablation is striking—raw images yield 80.2 percent Dice, the full pipeline 88.3 percent, and the boundary error metric HD95 falls from 12.6 pixels to 4.1 pixels.

The genuinely novel element sits at the skip connections, the channels through which a UNet passes encoder features directly to the decoder. In a standard UNet these connections are a liability in angiography: they carry catheter outlines, bone edges, and heart-shadow noise straight into the reconstruction path, weakening the recovery of thin vessels. The team replaced them with a Multi-Scale Adaptive Spatial Attention Gate, or MASAG, adapted from the MSA2Net architecture of Kolahi and colleagues and installed at all four encoder skip levels, with channel dimensions of 64, 64, 128, and 256. Each MASAG operates in four stages. It first extracts multi-scale features from the encoder signal using a depth-wise separable 3 by 3 convolution for fine vessel detail and a dilated convolution with rate 2 for wider structures such as curves and branches, fusing them with global decoder context aligned through global average pooling and upsampling. A Softmax-based spatial selection map then generates two attention weights that reweight the encoder and decoder features, allowing the two paths to compete per pixel for influence.

What follows is the two-way interaction that distinguishes MASAG from earlier attention modules like CBAM or the attention gates of Schlemper and colleagues. The reweighted encoder and decoder features cross-modulate each other—the encoder signal modulates the decoder guidance and vice versa—producing an interaction feature that captures joint vessel-relevant information from both directions. Finally, a channel attention vector computed with a reduction ratio of 8 recalibrates the output before it reaches the decoder. The ablation table makes the case quantitatively: within the identical ResNet50-UNet backbone, an attention gate achieves 85.4 percent Dice, CBAM 86.6 percent, and MASAG 88.3 percent. Removing MASAG entirely drops performance by roughly 8 to 10 Dice points, and equipping all four skip connections outperforms one, two, or three, with the four-skip configuration reaching the full 88.3 percent. Image-level paired statistical tests across all 134 out-of-fold predictions confirm the gains are significant, with Cohen’s d effect sizes of 1.55 or larger and p-values below 0.001 after Bonferroni correction.

Attention alone, however, cannot guarantee that the predicted vessel tree remains anatomically connected. Binary cross-entropy and Dice losses optimize pixel overlap and region similarity, but a prediction can score well on both while fragmenting the centreline into disconnected pieces—a fatal flaw for any downstream stenosis analysis that depends on tracing a continuous path along an artery. The team therefore trained with a hybrid loss combining three terms in equal weights: binary cross-entropy for pixel-level refinement, Dice for region overlap and class imbalance, and centreline Dice, or clDice, for topological connectivity. clDice, introduced by Shit and colleagues in 2021, compares the centreline structures of prediction and ground truth using a differentiable soft-skeletonization operation in which iterative min and max pooling with a 3 by 3 kernel approximate morphological erosion and dilation. Five iterations suffice to extract thin coronary skeletons at the 256 by 256 working resolution. When a vessel breaks, the skeleton fragments, both centreline-matching terms fall, and the loss rises—forcing the network to learn connectivity as a training objective rather than hoping for it as a by-product.

The topology-aware loss earns its place in the numbers. Adding clDice to a Dice-plus-BCE baseline lifts Dice from 85.9 to 88.3 percent, the clDice metric itself from 86.3 to 89.1 percent, HD95 from 7.4 to 4.1 pixels, and the Betti-0 error—a count of connected components in the prediction versus the ground truth, using 8-connectivity appropriate for thin diagonally oriented vessels—from 3.1 to 1.5. A sensitivity analysis varying the clDice loss weight from 0.1 to 5.0 shows performance stable across the range, peaking at the balanced equal-weight setting, so the choice is not a fragile hyperparameter. The team also reports the 95th percentile Hausdorff distance to capture boundary-level accuracy and penalize large local deviations, and they candidly note a limitation of Betti-0 itself: because it compares only component counts, a false merge and a false split can offset each other and leave the count unchanged, so they interpret it alongside overlap and boundary metrics rather than in isolation.

Training followed a two-phase schedule on an NVIDIA Tesla P100 GPU in the Kaggle environment using Keras-TensorFlow with Python 3.10. For the first 25 epochs the ResNet50 encoder stayed frozen while the MASAG gates and decoder learned stable features at a learning rate of 1 times 10 to the minus 4; the encoder was then fully unfrozen for another 25 epochs of joint fine-tuning, a strategy the ablation confirms outperforms both a permanently frozen encoder at 84.1 percent Dice and selective unfreezing of deeper stages at 86.5 percent. The complete model carries 37,397,633 parameters, requires roughly 32.5 GFLOPs per forward pass at 256 by 256 input, trains in about four minutes per fold, and infers in approximately 22 milliseconds per image—fast enough, in principle, for clinical workflows. Five-fold cross-validation, with four folds splitting 107 training and 27 validation images and one fold 108 and 26, produced the reported means and standard deviations, and the learning curves show training and validation losses declining in close alignment from roughly 0.90 to 0.01 and 0.88 to 0.02 respectively, with narrow shaded bands across folds indicating stable generalization without major overfitting.

The headline comparison is blunt: the standard UNet baseline reaches 74.0 percent Dice, UNet++ 74.7 percent, and a ResNet50-UNet without attention 76.1 percent, while the full proposed model attains 88.3 percent Dice, 79.1 percent Jaccard, 86.4 percent precision, and 90.5 percent recall. Against recently reported methods on DCA1—including FiTNet at 81.1 percent, EAG-SAE-ESTF at 80.6 percent, and CoorMa at 80.4 percent—the proposed model reports among the highest Dice, Jaccard, and recall values, though the authors are careful to state that these studies used different preprocessing and evaluation protocols, so the comparison is a reference rather than a strictly controlled head-to-head benchmark. Qualitative results, with true positives in green, false positives in red, and false negatives in blue, show the model maintaining a consistent balance between error types and suppressing non-vascular activations from catheters and bones without any separate removal step.

The authors are equally candid about the boundaries of the achievement. The evaluation rests on a single public dataset of 134 images with cross-validation folds rather than a separate test set, so performance may appear slightly better than it actually is, and patient-level independence between training and validation splits cannot be verified because the distributed images carry no patient identifiers. Thin distal vessel branches are sometimes missed, very low-contrast segments can blend into surrounding tissue, and unsharp masking and CLAHE may amplify background noise in some areas despite bilateral filtering and a moderate clip limit. Most tellingly, a zero-shot test on the external XCAD dataset of 126 images from a different imaging system—resized but not preprocessed—yielded a Dice of just 22.3 percent, a sobering measure of the domain gap between imaging systems and a reminder that cross-dataset adaptation remains future work. No clinical tasks such as stenosis quantification were evaluated, and the authors note that expert-annotated stenosis data, which DCA1 does not provide, would be required for blockage percentage analysis. Still, the structural basis is now in place: a segmentation that keeps the coronary tree connected, suppresses the visual clutter of the catheter lab, and runs in milliseconds. Whether that foundation survives contact with diverse hospital data from different scanners is the question the next phase of this research must answer.

Subject of Research: Artificial Intelligence

Article Title: Topology-preserving coronary vessel segmentation in X-ray angiography using ResNet50-MASAG-UNet

Article References: Sunilkumar, G., & Kumaresan, P. (2026). Topology-preserving coronary vessel segmentation in X-ray angiography using ResNet50-MASAG-UNet. Frontiers in Radiology, 6, Article 1949450. https://doi.org/10.3389/fradi.2026.1949450

Image Credits: AI Generated

DOI: 10.3389/fradi.2026.1949450

Keywords: coronary artery disease, X-ray coronary angiography, deep learning, vessel segmentation, attention gates, topology-preserving loss, ResNet50, medical imaging, Topology-preserving, coronary, vessel, segmentation

The clinical stakes are considerable. During stent-assisted treatment, a catheter is inserted through the femoral artery in the groin or the radial artery in the wrist and navigated through tortuous, branching vessels deep inside the brain. The stent must then be positioned so that its ends rest in straight sections of the vessel, providing stable anchoring while fully covering the aneurysm orifice. Misjudging the geometry can lead to stent migration, arterial occlusion, vessel damage, or stroke. Stent selection itself is a delicate balancing act: the device must match the diameter and length of the parent artery, accommodate the curvature of the vascular path, and leave adequate landing zones on both sides of the aneurysm neck. Traditional planning based on CT angiography, MR angiography, or digital subtraction angiography flattens this complex three-dimensional puzzle onto two dimensions, forcing clinicians to reconstruct the anatomy mentally at the very moment when errors matter most.

The Krakow team, led by Wojciech Chmiel together with Piotr Szwed, Joanna Kwiecień, Michał Turek and colleagues, built their platform around a workflow engine that automates the entire chain of image processing. Starting from volumetric medical images in NIfTI format, the system performs vessel segmentation using threshold-based binarization techniques such as Otsu’s method, cleans the resulting point clouds with the DBSCAN clustering algorithm to strip away outliers, and then computes the vessel’s medial axis, the invisible central spine along which a stent naturally wants to sit. The centerline extraction is particularly elegant: a distance-based potential field is evaluated for every interior point of the vessel, with the lowest potential values forming a valley along the vessel’s core, and the Dijkstra shortest-path algorithm then traces the optimal route between two operator-selected endpoints. Because the raw path follows a voxel grid and emerges as a staircase curve, a Savitzky-Golay filter smooths it into a anatomically plausible centerline without erasing genuine high-frequency curvature.

Once the centerline is established, the system computes a radial profile at each point, casting 120 rays spaced three degrees apart in a plane orthogonal to the axis and measuring the distance to the vessel wall along each one. These profiles act as a geometric fingerprint of the artery, allowing the software to pinpoint aneurysm necks and bifurcations with remarkable fidelity. At a spatial resolution of 0.05 millimeters per voxel, the smoothed profiles capture subtle variations in lumen diameter that would be invisible on a standard angiographic display. The operator then marks an attraction region on the centerline, a deliberately generous zone within which the stent simulation will unfold, and the processed mesh is exported to a format compatible with Microsoft’s HoloLens headset, complete with correctly oriented surface normal vectors that are essential for realistic shading and depth perception in stereoscopic rendering.

The heart of the system is its stent expansion simulation, which the authors describe as a computationally efficient, slice-based geometric approximation rather than a full biomechanical model. The stent is represented as a sequence of planar cross-sectional slices connected into a tubular triangular mesh. Each slice is iteratively expanded using cylindrical transformations, and when the expanding ring encounters the vessel wall, contact points freeze in place while the remaining points continue to move outward, deforming the ring to conform to the arterial geometry. These contact points are called support points, and the final slice border is interpolated through them using cubic splines. The researchers implemented four interaction modes, from rigid slice with rigid vessel to fully flexible combinations, and chose the flexible slice, rigid vessel configuration as the default because cerebral arteries do not dilate significantly during stent deployment. Restoring forces between neighboring mesh nodes smooth the vessel surface and prevent expansion artifacts, and stopping criteria based on slice asymmetry and area stagnation freeze slices that can no longer expand meaningfully.

What makes this approach genuinely practical for the operating room is speed. Full finite element simulations of stent deployment, while physically rigorous, are far too slow for real-time interaction. The Krakow team’s geometric approximation produces plausible, collision-aware stent configurations in seconds, generating what they call a solution space: a discrete set of candidate stent positions from which the clinician can select the configuration that best matches a specific device’s nominal diameter and foreshortening behavior. The system can even test microcatheter navigability, using rigid slice expansion to determine whether catheters with outer diameters between roughly 0.5 and 0.93 millimeters can physically pass through a given vascular segment. If slices freeze along the path, that signals a passage too tight for the chosen device, an early warning that could redirect the entire treatment strategy before the patient is ever on the table.

The augmented reality layer is where the system becomes genuinely striking. Through the HoloLens application, the operator sees a magnified, stereoscopic rendering of the patient’s vessel floating in the operating room, with semi-transparent stent models that can be picked from a virtual panel, moved into the artery, rotated, and snapped into alignment along the centerline, all through hand gestures detected by the headset’s cameras. Because touching a keyboard or mouse in a sterile surgical environment is impractical, the gesture subsystem was built on the Mixed Reality Toolkit 2 API and works even while the operator wears rubber surgical gloves. Perhaps most remarkably, the vessel model is scaled and calibrated so that the clinician can literally walk inside the rendered artery, inspecting the stent surface and its apposition to the vessel wall from the inside, an perspective no flat screen can provide.

Meeting the HoloLens’s demanding performance constraints required serious engineering discipline. The headset refreshes at 60 to 72 frames per second, leaving a render budget of under 16.6 milliseconds per frame to avoid inducing motion sickness, and stereoscopic rendering doubles the workload by requiring separate viewports for each eye. The team kept active geometry below 150,000 polygons per frame, minimized draw calls through batching, employed mobile-optimized shaders, and used Single-Pass Instanced Rendering to process both eye viewports in a single GPU execution. Late Stage Reprojection dynamically adjusts each frame’s focal point just before display, compensating for head-tracking latency and preventing parallax errors between the virtual models and the physical environment. These optimizations ensure that the immersive experience remains stable and comfortable even during prolonged planning sessions.

Beyond its intraoperative role, the platform doubles as a training and education tool. Every action, parameter choice, and correction made by a specialist during a workflow is recorded in a procedure repository, creating an audit trail that can be replayed for educational purposes or, subject to appropriate governance, used to train future artificial intelligence algorithms. This human-in-the-loop feedback cycle connects clinical expertise directly to algorithm development, addressing a well-known weakness in medical AI: the scarcity of large, expertly annotated datasets. The system’s workflow architecture is deliberately modular, meaning that today’s reference implementations of segmentation and simulation can be swapped for more advanced methods, including deep-learning approaches, without rebuilding the user interface or retraining clinicians on a new platform.

The authors are candid about limitations. The geometric stent simulation has not yet been quantitatively validated against benchtop deployment experiments, manufacturer data, or full finite element models, and it should not be regarded as more physically accurate than rigorous biomechanical approaches. Its advantages lie elsewhere: interactive speed, direct integration with the augmented reality workflow, and the ability to explore many candidate configurations in the time a single FEM run would take. Augmented reality headsets also impose inherent constraints on field of view and graphical resolution, though the team argues that vision-based tracking and semi-transparent rendering preserve sufficient localization precision even at high magnification. Looking forward, the researchers envision AI algorithms automating critical stages of image processing and generating comprehensive libraries of interventional scenarios. For now, the system stands as a compelling demonstration that 3D imaging, fast simulation, and immersive visualization can converge into a single tool that may make one of neurosurgery’s most delicate procedures measurably safer, more precise, and easier to teach.

Subject of Research: Augmented reality and 3D imaging support for stent-assisted treatment of cerebral aneurysms

Article Title: A Visual Support System for Stent-Assisted Treatment of Cerebral Aneurysms Using 3D Imaging and Augmented Reality

Article References: Chmiel, W., Szwed, P., Kwiecień, J., Turek, M., Jȩdrusik, S., Pałka, D., Waśniewski, M., Kadłuczka, P., Gumiela, D., & Zeja, A. (2026). A Visual Support System for Stent-Assisted Treatment of Cerebral Aneurysms Using 3D Imaging and Augmented Reality. Neuroinformatics, 24(4), Article 65. https://doi.org/10.1007/s12021-026-09812-2

Image Credits: AI Generated

DOI: 10.1007/s12021-026-09812-2

Keywords: cerebral aneurysm, augmented reality, stent-assisted coiling, 3D imaging, medical simulation, HoloLens, neurointerventional surgery, vessel segmentation, medial axis, workflow engine, digital health, Neuroinformatics