International Journal of Computer Assisted Radiology and Surgery最新文献

Purpose: Dysphagia is the inability or difficulty to swallow normally. Standard procedures for diagnosing the exact disease are, among others, X-ray videofluoroscopy, manometry and impedance examinations, usually performed consecutively. In order to gain more insights, ongoing research is aiming to collect these different modalities at the same time, with the goal to present them in a joint visualization. One idea to create a combined view is the projection of the manometry and impedance values onto the right location in the X-ray images. This requires to identify the exact sensor locations in the images.

Methods: This work gives an overview of the challenges associated with the sensor detection task and proposes a robust approach to detect the sensors in X-ray image sequences, ultimately allowing to project the manometry and impedance values onto the right location in the images.

Results: The developed sensor detection approach is evaluated on a total of 14 sequences from different patients, achieving a F1-score of 86.36%. To demonstrate the robustness of the approach, another study is performed by adding different levels of noise to the images, with the performance of our sensor detection method only slightly decreasing in these scenarios. This robust sensor detection provides the basis to accurately project manometry and impedance values onto the images, allowing to create a multimodal visualization of the swallow process. The resulting visualizations are evaluated qualitatively by domain experts, indicating a great benefit of this proposed fused visualization approach.

Conclusion: Using our preprocessing and sensor detection method, we show that the sensor detection task can be successfully approached with high accuracy. This allows to create a novel, multimodal visualization of esophageal motility, helping to provide more insights into swallow disorders of patients.

Purpose: The pathomorphology of Legg-Calvé-Perthes disease (LCPD) is a key contributor to poor long-term outcomes such as hip pain, femoroacetabular impingement, and early-onset osteoarthritis. Plain radiographs, commonly used for research and in the clinic, cannot accurately represent the full extent of LCPD deformity. The purpose of this study was to develop and evaluate a methodological framework for three-dimensional (3D) statistical shape modeling (SSM) of the proximal femur in LCPD.

Methods: We developed a framework consisting of three core steps: segmentation, surface mesh preparation, and particle-based correspondence. The framework aims to address challenges in modeling this rare condition, characterized by highly heterogeneous deformities across a wide age range and small sample sizes. We evaluated this framework by producing a SSM from clinical magnetic resonance images of 13 proximal femurs with LCPD deformity from 11 patients between the ages of six and 12 years.

Results: After removing differences in scale and pose, the dominant shape modes described morphological features characteristic of LCPD, including a broad and flat femoral head, high-riding greater trochanter, and reduced neck-shaft angle. The first four shape modes were chosen for the evaluation of the model's performance, together describing 87.5% of the overall cohort variance. The SSM was generalizable to unfamiliar examples with an average point-to-point reconstruction error below 1mm. We observed strong Spearman rank correlations (up to 0.79) between some shape modes, 3D measurements of femoral head asphericity, and clinical radiographic metrics.

Conclusion: In this study, we present a framework, based on SSM, for the objective description of LCPD deformity in three dimensions. Our methods can accurately describe overall shape variation using a small number of parameters, and are a step toward a widely accepted, objective 3D quantification of LCPD deformity.

Purpose: Teleoperated Interventional Robotic systems (TIRs) are developed to reduce radiation exposure and physical stress of the physicians and enhance device manipulation accuracy and stability. Nevertheless, TIRs are not widely adopted, partly due to the lack of intuitive control interfaces. Current TIR interfaces like joysticks, keyboards, and touchscreens differ significantly from traditional manual techniques, resulting in a shallow, longer learning curve. To this end, this research introduces a novel control mechanism for intuitive operation and seamless adoption of TIRs.

Methods: An off-the-shelf medical torque device augmented with a micro-electromagnetic tracker was proposed as the control interface to preserve the tactile sensation and muscle memory integral to interventionalists' proficiency. The control inputs to drive the TIR were extracted via real-time motion mapping of the interface. To verify the efficacy of the proposed control mechanism to accurately operate the TIR, evaluation experiments using industrial grade encoders were conducted.

Results: A mean tracking error of 0.32 ± 0.12 mm in linear and 0.54 ± 0.07° in angular direction were achieved. The time lag in tracking was found to be 125 ms on average using pade approximation. Ergonomically, the developed control interface is 3.5 mm diametrically larger, and 4.5 g. heavier compared to traditional torque devices.

Conclusion: With uncanny resemblance to traditional torque devices while maintaining results comparable to state-of-the-art commercially available TIRs, this research successfully provides an intuitive control interface for potential wider clinical adoption of robot-assisted interventions.

Purpose: This study investigates the application of Radiomic features within graph neural networks (GNNs) for the classification of multiple-epitope-ligand cartography (MELC) pathology samples. It aims to enhance the diagnosis of often misdiagnosed skin diseases such as eczema, lymphoma, and melanoma. The novel contribution lies in integrating Radiomic features with GNNs and comparing their efficacy against traditional multi-stain profiles.

Methods: We utilized GNNs to process multiple pathological slides as cell-level graphs, comparing their performance with XGBoost and Random Forest classifiers. The analysis included two feature types: multi-stain profiles and Radiomic features. Dimensionality reduction techniques such as UMAP and t-SNE were applied to optimize the feature space, and graph connectivity was based on spatial and feature closeness.

Results: Integrating Radiomic features into spatially connected graphs significantly improved classification accuracy over traditional models. The application of UMAP further enhanced the performance of GNNs, particularly in classifying diseases with similar pathological features. The GNN model outperformed baseline methods, demonstrating its robustness in handling complex histopathological data.

Conclusion: Radiomic features processed through GNNs show significant promise for multi-disease classification, improving diagnostic accuracy. This study's findings suggest that integrating advanced imaging analysis with graph-based modeling can lead to better diagnostic tools. Future research should expand these methods to a wider range of diseases to validate their generalizability and effectiveness.

Purpose: Proper visualization and interaction with complex anatomical data can improve understanding, allowing for more intuitive surgical planning. The goal of our work was to study what the most intuitive yet practical platforms for interacting with 3D medical data are in the context of surgical planning.

Methods: We compared planning using a monitor and mouse, a monitor with a haptic device, and an augmented reality (AR) head-mounted display which uses a gesture-based interaction. To determine the most intuitive system, two user studies, one with novices and one with experts, were conducted. The studies involved planning of three scenarios: (1) heart valve repair, (2) hip tumor resection, and (3) pedicle screw placement. Task completion time, NASA Task Load Index and system-specific questionnaires were used for the evaluation.

Results: Both novices and experts preferred the AR system for pedicle screw placement. Novices preferred the haptic system for hip tumor planning, while experts preferred the mouse and keyboard. In the case of heart valve planning, novices preferred the AR system but there was no clear preference for experts. Both groups reported that AR provides the best spatial depth perception.

Conclusion: The results of the user studies suggest that different surgical cases may benefit from varying interaction and visualization methods. For example, for planning surgeries with implants and instrumentations, mixed reality could provide better 3D spatial perception, whereas using landmarks for delineating specific targets may be more effective using a traditional 2D interface.

Purpose: Clinical needle insertion into tissue, commonly assisted by 2D ultrasound imaging for real-time navigation, faces the challenge of precise needle and probe alignment to reduce out-of-plane movement. Recent studies investigate 3D ultrasound imaging together with deep learning to overcome this problem, focusing on acquiring high-resolution images to create optimal conditions for needle tip detection. However, high-resolution also requires a lot of time for image acquisition and processing, which limits the real-time capability. Therefore, we aim to maximize the US volume rate with the trade-off of low image resolution. We propose a deep learning approach to directly extract the 3D needle tip position from sparsely sampled US volumes.

Methods: We design an experimental setup with a robot inserting a needle into water and chicken liver tissue. In contrast to manual annotation, we assess the needle tip position from the known robot pose. During insertion, we acquire a large data set of low-resolution volumes using a 16  $\times$  16 element matrix transducer with a volume rate of 4 Hz. We compare the performance of our deep learning approach with conventional needle segmentation.

Results: Our experiments in water and liver show that deep learning outperforms the conventional approach while achieving sub-millimeter accuracy. We achieve mean position errors of 0.54 mm in water and 1.54 mm in liver for deep learning.

Conclusion: Our study underlines the strengths of deep learning to predict the 3D needle positions from low-resolution ultrasound volumes. This is an important milestone for real-time needle navigation, simplifying the alignment of needle and ultrasound probe and enabling a 3D motion analysis.

Purpose: We introduce a novel approach for bronchoscopic navigation that leverages neural radiance fields (NeRF) to passively locate the endoscope solely from bronchoscopic images. This approach aims to overcome the limitations and challenges of current bronchoscopic navigation tools that rely on external infrastructures or require active adjustment of the bronchoscope.

Methods: To address the challenges, we leverage NeRF for bronchoscopic navigation, enabling passive endoscope localization from bronchoscopic images. We develop a two-stage pipeline: offline training using preoperative data and online passive pose estimation during surgery. To enhance performance, we employ Anderson acceleration and incorporate semantic appearance transfer to deal with the sim-to-real gap between training and inference stages.

Results: We assessed the viability of our approach by conducting tests on virtual bronchscopic images and a physical phantom against the SLAM-based methods. The average rotation error in our virtual dataset is about 3.18 $_{}^{\circ }$ and the translation error is around 4.95 mm. On the physical phantom test, the average rotation and translation error are approximately 5.14 $_{}^{\circ }$ and 13.12 mm.

Conclusion: Our NeRF-based bronchoscopic navigation method eliminates reliance on external infrastructures and active adjustments, offering promising advancements in bronchoscopic navigation. Experimental validation on simulation and real-world phantom models demonstrates its efficacy in addressing challenges like low texture and challenging lighting conditions.

Purpose: Most recently transformer models became the state of the art in various medical image segmentation tasks and challenges, outperforming most of the conventional deep learning approaches. Picking up on that trend, this study aims at applying various transformer models to the highly challenging task of colorectal cancer (CRC) segmentation in CT imaging and assessing how they hold up to the current state-of-the-art convolutional neural network (CNN), the nnUnet. Furthermore, we wanted to investigate the impact of the network size on the resulting accuracies, since transformer models tend to be significantly larger than conventional network architectures.

Methods: For this purpose, six different transformer models, with specific architectural advancements and network sizes were implemented alongside the aforementioned nnUnet and were applied to the CRC segmentation task of the medical segmentation decathlon.

Results: The best results were achieved with the Swin-UNETR, D-Former, and VT-Unet, each transformer models, with a Dice similarity coefficient (DSC) of 0.60, 0.59 and 0.59, respectively. Therefore, the current state-of-the-art CNN, the nnUnet could be outperformed by transformer architectures regarding this task. Furthermore, a comparison with the inter-observer variability (IOV) of approx. 0.64 DSC indicates almost expert-level accuracy. The comparatively low IOV emphasizes the complexity and challenge of CRC segmentation, as well as indicating limitations regarding the achievable segmentation accuracy.

Conclusion: As a result of this study, transformer models underline their current upward trend in producing state-of-the-art results also for the challenging task of CRC segmentation. However, with ever smaller advances in total accuracies, as demonstrated in this study by the on par performances of multiple network variants, other advantages like efficiency, low computation demands, or ease of adaption to new tasks become more and more relevant.

Purpose: Video-based intra-abdominal instrument tracking for laparoscopic surgeries is a common research area. However, the tracking can only be done with instruments that are actually visible in the laparoscopic image. By using extra-abdominal cameras to detect trocars and classify their occupancy state, additional information about the instrument location, whether an instrument is still in the abdomen or not, can be obtained. This can enhance laparoscopic workflow understanding and enrich already existing intra-abdominal solutions.

Methods: A data set of four laparoscopic surgeries recorded with two time-synchronized extra-abdominal 2D cameras was generated. The preprocessed and annotated data were used to train a deep learning-based network architecture consisting of a trocar detection, a centroid tracker and a temporal model to provide the occupancy state of all trocars during the surgery.

Results: The trocar detection model achieves an F1 score of $95.06\pm 0.88\%$ . The prediction of the occupancy state yields an F1 score of $89.29\pm 5.29\%$ , providing a first step towards enhanced surgical workflow understanding.

Conclusion: The current method shows promising results for the extra-abdominal tracking of trocars and their occupancy state. Future advancements include the enlargement of the data set and incorporation of intra-abdominal imaging to facilitate accurate assignment of instruments to trocars.

Purpose: The current study explores the application of 3D U-Net architectures combined with Inception and ResNet modules for precise lung nodule detection through deep learning-based segmentation technique. This investigation is motivated by the objective of developing a Computer-Aided Diagnosis (CAD) system for effective diagnosis and prognostication of lung nodules in clinical settings.

Methods: The proposed method trained four different 3D U-Net models on the retrospective dataset obtained from AIIMS Delhi. To augment the training dataset, affine transformations and intensity transforms were utilized. Preprocessing steps included CT scan voxel resampling, intensity normalization, and lung parenchyma segmentation. Model optimization utilized a hybrid loss function that combined Dice Loss and Focal Loss. The model performance of all four 3D U-Nets was evaluated patient-wise using dice coefficient and Jaccard coefficient, then averaged to obtain the average volumetric dice coefficient (DSC_avg) and average Jaccard coefficient (IoU_avg) on a test dataset comprising 53 CT scans. Additionally, an ensemble approach (Model-V) was utilized featuring 3D U-Net (Model-I), ResNet (Model-II), and Inception (Model-III) 3D U-Net architectures, combined with two distinct patch sizes for further investigation.

Results: The ensemble of models obtained the highest DSC_avg of 0.84 ± 0.05 and IoU_avg of 0.74 ± 0.06 on the test dataset, compared against individual models. It mitigated false positives, overestimations, and underestimations observed in individual U-Net models. Moreover, the ensemble of models reduced average false positives per scan in the test dataset (1.57 nodules/scan) compared to individual models (2.69-3.39 nodules/scan).

Conclusions: The suggested ensemble approach presents a strong and effective strategy for automatically detecting and delineating lung nodules, potentially aiding CAD systems in clinical settings. This approach could assist radiologists in laborious and meticulous lung nodule detection tasks in CT scans, improving lung cancer diagnosis and treatment planning.