International Journal of Computer Assisted Radiology and Surgery最新文献

Purpose: Low-light endoscopic images often lack contrast and clarity, obscuring anatomical details and reducing diagnostic accuracy. This study develops a method to enhance image brightness and visibility, enabling clearer visualization of critical structures to support precise medical diagnoses and improve patient outcomes.

Methods: To specifically address nonuniform illumination, we propose BrightVAE, a model that uses a dual-receptive-field architecture to decouple global brightness correction from local texture preservation. Integrated attention-based modules (Attencoder and Attenquant) explicitly target and amplify underexposed regions while preventing over-saturation, thereby recovering human-evaluable details in shadowed areas. The model was trained and tested on a public endoscopic dataset, and its performance was evaluated against other techniques using quality metrics.

Results: The model outperformed alternatives, improving PSNR by 3.252 units, structural detail by 0.045, and perceptual quality by 0.014 compared to the best model before us, achieving a PSNR of 30.576, SSIM of 0.879, and LPIPS of 0.133, ensuring superior visibility of shadowed regions.

Conclusion: This approach advances endoscopic imaging by delivering sharper, reliable images, enhancing diagnostic precision in clinical practice. Improved visualization supports better detection of abnormalities, potentially leading to more effective treatment decisions and enhanced patient care.

Purpose: This study aims to evaluate the hemodynamic effect of tilting after implantation of a bioprosthetic heart valve in an enlarged aortic annulus using computational fluid dynamics (CFD). The objective is to define the cost of excessive enlargement in transprosthetical flow and identify the optimal prosthesis size and implantation angle to reduce thrombosis and early degeneration.

Methods: Virtual implantation of a 23-mm bioprosthetic model was conducted in a prespecified virtual aortic annulus of 23 mm. CFD simulations were conducted to analyze the hemodynamic parameters, including wall shear stress and turbulent kinetic energy, for bioprosthetic heart valves of the 23 mm (perpendicular implantation), 25 mm, and 27 mm (tilted implantation at different angles) after virtual annular enlargement. The simulations utilized transient flow models and mesh convergence studies, to ensure numerical accuracy and clinical relevance.

Results: The 23-mm valve implanted without annular enlargement and in an aligned fashion to the annulus exhibited higher wall shear stress (WSS) and shear stress (SS) values than the 25-mm valve implanted at 12° tilted position and better hemodynamics than the 27-mm valve at 25° tilted position. The 25-mm valve after annular enlargement and implantation at 12° tilted position achieved the best hemodynamic performance together with a 23% and 7% reduction of WSS and SS as compared to the 23-mm valve.

Conclusions: Slight oversizing and tilting after annular enlargement yields the best performance of an aortic bioprosthetic valve with lowest WSS and balanced turbulence.

Purpose: Accurate 3D reconstruction from endoscopic videos is crucial for advancing computer-assisted minimally invasive surgery. However, existing approaches struggle with dynamic surgical scenes where instrument occlusions cause significant reconstruction artifacts. Although 3D Gaussian Splatting (3DGS) enables rapid reconstruction, it often suffers from incomplete surface recovery due to occlusion-induced missing regions and error propagation from suboptimal initial point clouds during radiance field optimization. This study aims to enhance reconstruction accuracy in dynamically occluded surgical environments.

Methods: We propose a diffusion-guided Gaussian Splatting (DiGS) framework comprising two key components: (1) a diffusion-guided surface completion network that incorporates surgical scene priors to restore high-fidelity textures in occluded regions, improving surface completeness; and (2) a lightweight annealed smoothing mechanism designed to mitigate endoscope motion estimation errors, ensuring temporal coherence during continuous frame interpolation and stabilizing radiance field optimization.

Results: Extensive experiments on the EndoNeRF and StereoMIS datasets demonstrate the superiority of DiGS over state-of-the-art baselines. On EndoNeRF, DiGS achieves a 61.75% improvement in LPIPS, indicating stronger perceptual alignment in dynamically occluded scenes. On StereoMIS, DiGS delivers an 7.03% PSNR gain and a 40.79% LPIPS improvement, along with consistently higher SSIM scores confirming superior preservation of structural details.

Conclusion: The proposed DiGS framework effectively addresses the challenges of dynamic occlusions and motion-induced errors in surgical scene reconstruction, producing more accurate and temporally coherent 3D models. The code is publicly available at https://github.com/IGSResearch/DiGS .

Purpose: This paper proposes a left-right (LR) relationship-aware classification model for 3D volumetric images (3D volume). Bilateral symmetry (LR relationship) is an essential property of the human body that can be used to detect abnormalities and understand anatomical structures. Checking the difference and similarity between the left and right anatomical structures is very important in diagnosis. We propose an LR relationship-aware classification model of 3D volume.

Methods: The proposed model employs an image feature extraction process from LR symmetric positions of human anatomy from 3D volume. Due to variations in body position and individual anatomical structure, small positional gaps among LR corresponding anatomical structures can be observed in medical images. We developed a multi-shift symmetric feature extraction module to accommodate such positional gaps.

Results: The model was applied to 3D volume classification tasks of the lung and brain. From experimental results, the proposed model achieved superior performances both in the lung and brain classification tasks compared to the previous models. The result indicates that the proposed model has generalized performance in classifying anatomical structures with bilateral symmetric or semi-symmetric structures.

Conclusion: We proposed the LR relationship-aware classification model of 3D volume. The proposed model effectively extracts image features from LR symmetric positions. The multi-shift symmetric feature extraction module was employed to accommodate small positional gaps among LR corresponding positions. The experimental results of 3D volume classification tasks of the lung and brain showed that the proposed method achieved superior performances compared to the previous models. Our code is available at https://github.com/modafone/lr3dvolumeclassification .

Purpose: Positioning of an osteosynthesis plate is a key step in the preoperative 3D planning processes for the design of patient-specific guides. This step requires considerable time and expertise. To increase 3D planning efficiency, this study aims to develop an automated plate positioning algorithm.

Methods: A robust algorithm was developed to optimize osteosynthesis plate positioning on the distal radius, using STL properties and anatomical landmarks. The algorithm involved alignment, landmark detection, initial placement, and final optimization. Retrospective data of 34 planned radii and corresponding plate positions, including decimated and refined mesh versions, were used to compare algorithm output to manual placement based on runtime, Hausdorff distance, translation, and rotation (mean ± SD, 95% CI), thereby assessing robustness across different mesh resolutions.

Results: The average run time for the algorithm was 18.3 ± 16.8 s (95% CI 12.4-24.1 s) compared to a manual placement time of 12.45 ± 4.56 min (single expert, n = 10, 95% CI 9.22-16.28 min). The mean unpaired maximum Hausdorff distance between manual and algorithm placements was 5.5 ± 2.5 mm (95% CI 4.6-6.4 mm). The mean rotation and translation differences were 4.9 ± 3.2° (95% CI 3.8-6.0°) and 3.3 ± 1.7 mm (95% CI 2.8-3.9 mm), respectively.

Conclusion: In conclusion, while some manual adjustment remains necessary, the algorithm aids in reducing planning time and offers a modular, generalizable framework adaptable to other osteotomy-plate procedures, supporting clinical 3D planning.

Purpose: In surgical skill development, a trainee's goal is to move through the learning curve and achieve expert performance. Goal directed training, with expert performance as the goal, can facilitate skill development; however, there are currently few methods available to encode expert-like simulation performance into learning strategies that can be practiced independently.

Methods: We propose a novel method of surgical simulation skill analysis through segmenting and evaluating kinematic data with instantaneous screw axes (ISA) theory and K-means clustering. In ISA, single degree of freedom (DOF) tasks can be represented as displacements about a single screw axis; however, we propose extending this method to more complex tasks defining them with clusters of similar ISAs in the surgical environment, decomposing them into a sequence of 1DOF movements. In this paper, we present an ISA algorithm and apply it to surgeon manipulator poses across fourteen suturing and knot-tying gestures obtained from the JIGSAWS surgical dataset. We also apply this method to entire simulated suturing demonstrations across a 6-month training period from the BGU-SKILLS dataset. We implemented K-means clustering to segment these movements into sub-gestures. We hypothesize that individuals with greater levels of expertise should exhibit more concise actions with minimal extraneous movement; therefore, fewer clusters should be required to decompose their simulation performance.

Results: Our ISA algorithm was applied to 1136 gestures from ten surgeons across three skill levels and 324 unsegmented demonstrations collected from 18 surgical residents over a training period of 6 months. We performed a Kruskal-Wallis analysis with a Dunn-Sidak post-hoc test on the number of ISA clusters required to decompose each gesture. We found that highly task-constrained gestures required significantly fewer numbers of clusters for expert and/or intermediate groups when compared to novices on suturing tasks only.

Conclusion: Our results suggest that this method can be used to identify task-constrained gestures within independently performed suturing surgical simulations and classify them into higher skill and lower skill sets. This analysis can also provide geometric feedback on performed gestures vs expert gestures, providing personalized automated performance analysis for surgical trainees leading to personalized educational training.

Purpose: Depth estimation from stereoscopic laparoscopic videos is of vital importance in computer-assisted intervention due to its potential for downstream tasks in laparoscopic surgical navigation. Previous works mostly focus on depth estimation from static frames, while temporal information in stereoscopic laparoscopic videos is largely ignored.

Methods: A spatiotemporal swin (ST-Swin) transformer-based network, referred to as ST-Swin TransNet, is proposed for depth estimation in stereoscopic surgical videos. Built upon a symmetric encoder-decoder architecture consisting of 12 ST-Swin blocks, ST-Swin TransNet extracts spatiotemporal features for efficient and accurate depth estimation, where the ST-Swin blocks are designed to capture spatiotemporal information from stereo video sequences via self-attention mechanism. Given binocular laparoscopic videos, ST-Swin TransNet exploits hierarchical spatiotemporal features to predict disparity maps.

Results: Comprehensive experiments are conducted on two typical yet challenging public datasets to evaluate the performance of the proposed method. We additionally demonstrate the feasibility of applying ST-Swin TransNet to video see-through augmented reality (VST-AR) navigation in laparoscopic surgery. Our method achieved a mean absolute depth error (mADE) of 3.33 mm in depth estimation and a mean absolute distance (mAD) of 1.07 mm in VST-AR navigation.

Conclusion: A spatiotemporal swin transformer-based network for self-supervised depth estimation in binocular laparoscopic surgical videos was developed. Results from the comprehensive experiments demonstrate the superior performance of the proposed method over the state-of-the-art methods.

Purpose Surgical gestures are fundamental components of surgical procedures, encompassing actions such as cutting, suturing, and knot-tying. Gesture recognition plays a pivotal role in the automatic analysis of surgical data. Although recent advancements have improved surgical gesture recognition, much of the existing research relies on simulations or minimally invasive surgery data, failing to capture the complexities of open surgery. In this study, we introduce and employ a new open surgery dataset focused on closing incisions after saphenous vein harvesting. Methods Our goal is to improve gesture recognition accuracy by incorporating tool pose estimation and 3D hand pose predictions of surgeons. We employ MS-TCN++  and LTContext  for gesture recognition, and further enhance performance through an ensemble of models using different modalities-video, tool pose, and hand pose data.Results The results reveal that using an ensemble model combining all three modalities provides a substantial improvement over video-only approaches, leading to statistically significant gains across multiple evaluation metrics. We further demonstrate that the model can rely solely on hand and tool poses, completely discarding the video input, while still achieving comparable performance. Additionally, we show that an ensemble model using only hand and tool poses produces results that are either: statistically significantly better than using video alone, or not statistically significantly different.Conclusion This study highlights the effectiveness of integrating multimodal data for surgical gesture recognition. By combining video, hand pose, and tool pose information, our approach achieves higher accuracy and robustness compared to video-only methods. Moreover, the comparable performance of pose-only models suggests a promising, privacy-preserving alternative for surgical data analysis.

Purpose: To discuss the economic and environmental implications of implementing large language models (LLMs) in radiology, highlighting both their transformative potential and the challenges they pose for equitable and sustainable adoption.

Methods: Current trends in AI investment, infrastructure requirements, operational costs, and environmental impact associated with LLMs are analyzed, highlighting the specific challenges of integrating LLMs into radiological workflows, including data privacy, regulatory compliance, and cost barriers for healthcare institutions. The analysis also considers the costs of model validation, maintenance, and updates, as well as investments in system integration, staff training, and cybersecurity for clinical implementation.

Results: LLMs have revolutionized natural language processing and offer promising applications in radiology, such as improved diagnostic support and workflow optimization. However, their deployment involves substantial financial and environmental costs. Training and operating these models require high-performance computing infrastructure, significant energy consumption, and large volumes of annotated data. Water usage and CO₂ emissions from data centers further raise sustainability concerns, while ongoing operational costs add to the financial burden. Subscription fees and per-query pricing may restrict access for smaller institutions, widening existing inequalities.

Conclusion: While LLMs offer significant benefits for radiology, their high economic and environmental costs present challenges to widespread and equitable adoption. Responsible use, sustainable practices, and policy frameworks are essential to ensure that AI-driven innovations do not exacerbate existing disparities in healthcare access and quality.