Search

Introduction: 3D surface scan-based diagnosis of craniosynostosis is a promising radiation-free alternative to traditional diagnosis using computed tomography. The cranial index (CI) and the cranial vault asymmetry index (CVAI) are well-established clinical parameters that are widely used. However, they also have the benefit of being easily adaptable for automatic diagnosis without the need of extensive preprocessing. Methods: We propose a multi-height-based classification approach that uses CI and CVAI in different height layers and compare it to the initial approach using only one layer. We use ten-fold cross-validation and test seven different classifiers. The dataset of 504 patients consists of three types of craniosynostosis and a control group consisting of healthy and non-synostotic subjects. Results: The multi-height-based approach improved classification for all classifiers. The k-nearest neighbors classifier scored best with a mean accuracy of 89% and a mean F1-score of 0.75. Conclusion: Taking height into account is beneficial for the classification. Based on accepted and widely used clinical parameters, this might be a step towards an easy-to-understand and transparent classification approach for both physicians and patients.

The application of machine learning approaches in medical technology is gaining more and more attention. Due to the high restrictions for collecting intraoperative patient data, synthetic data is increasingly used to support the training of artificial neural networks. We present a pipeline to create a statistical shape model (SSM) using 28 segmented clinical liver CT scans. Our pipeline consists of four steps: data preprocessing, rigid alignment, template morphing, and statistical modeling. We compared two different template morphing approaches: Laplace-Beltrami-regularized projection (LBRP) and nonrigid iterative closest points translational (N-ICP-T) and evaluated both morphing approaches and their corresponding shape model performance using six metrics. LBRP achieved a smaller mean vertex-to-nearest-neighbor distances (2.486±0.897 mm) than N-ICP-T (5.559±2.413 mm). Generalization and specificity errors for LBRP were consistently lower than those of N-ICP-T. The first principal components of the SSM showed realistic anatomical variations. The performance of the SSM was comparable to a state-of-the-art model.

Optical Coherence Tomography (OCT) is a standard imaging procedure in ophthalmology. OCT Angiography is a promising extension, allowing for fast and non-invasive imaging of the retinal vasculature analyzing multiple OCT scans at the same place. Local variance is examined and highlighted. Despite its introduction in the clinic, unanswered questions remain when it comes to signal generation. Multiphase fluids like intralipid, milk-water solutions and human blood cells were applied in phantom studies shedding light on some of the mechanisms. The use of hydrogel beads allows for the generation of alternative blood models for OCT and OCT Angiography. Beads were produced in Hannover, their size was measured and their long term stability was assessed. Then, beads were shipped to Karlsruhe, where OCT imaging resulted in first insights. The hydrogel acts as a diffusion barrier, which enables a clear distinction of bead and fluid when scattering particles were added. Further on, the scattering medium below the bead showed increased signal intensity. We conclude that the inside of the bead structure shows enhanced transmission compared to the plasma substitute with dissolved TiO2 surrounding it. Beads were found clumped and deformed after shipping, an issue to be addressed in further investigations. Nevertheless, hydrogel beads are promising as a blood model for OCT Angiography investigations, offering tunable optical parameters within the blood substitute solution.

During cerebral revascularization surgeries, blood flow values help surgeons to monitor the quality of the procedure, e.g., to avoid cerebral hyperperfusion syndrome due to excessively enhanced perfusion. The state-of-the-art technique is the ultrasonic flow probe that has to be placed around the blood vessel. This causes contact between probe and vessel, which, in the worst case, leads to rupture. The recently developed intraoperative indocyanine green (ICG) Quantitative Fluorescence Angiography (QFA) is an alternative technique that overcomes this risk. However, it has been shown by the developer that the calculated flow has deviations. After determining the bolus transit time as the most sensitive parameter in flow calculation, we propose a new two-step uncertainty reduction method for flow calculation. The first step is to generate more data in each measurement that results in functions of the parameters. Noise can then be reduced in a second step. Two methods for this step are compared. The first method fits the model for each parameter function separately and calculates flow from models, while the second one fits multiple parameter functions together. The latter method is proven to perform best by in silico tests. Besides, this method reduces the deviation of flow comparing to original QFA as expected. Our approach can be generally used in all QFA applications using two-point theory. Further development is possible if number of dimensions of the achieved parameter data are broadened that results in even more data for processing in the second step.

Craniosynostosis is a congenital disease characterized by the premature closure of one or multiple sutures of the infant’s skull. For diagnosis, 3D photogrammetric scans are a radiation-free alternative to computed tomography. However, data is only sparsely available and the role of data augmentation for the classification of craniosynostosis has not yet been analyzed. In this work, we use a 2D distance map representation of the infants’ heads with a convolutional-neural-network-based classifier and employ a generative adversarial network (GAN) for data augmentation. We simulate two data scarcity scenarios with 15% and 10% training data and test the influence of different degrees of added synthetic data and balancing underrepresented classes. We used total accuracy and F1-score as a metric to evaluate the final classifiers. For 15% training data, the GAN-augmented dataset showed an increased F1-score up to 0.1 and classification accuracy up to 3 %. For 10% training data, both metrics decreased. We present a deep convolutional GAN capable of creating synthetic data for the classification of craniosynostosis. Using a moderate amount of synthetic data using a GAN showed slightly better performance, but had little effect overall. The simulated scarcity scenario of 10% training data may have limited the model’s ability to learn the underlying data distribution.

In laparoscopic liver surgery (LLS) image-guided navigation systems could support the surgeon by providing subsurface information such as the positions of tumors and vessels. For this purpose, one option is to perform a registration of preoperative 3D data and 3D surface patches reconstructed from laparoscopic images. Part of an automatic 3D registration pipeline is the feature description, which takes into account various geometric and spatial information. Since there is no leading feature descriptor in the field of LLS, two feature descriptors are compared in this paper: The Fast Point Feature Histogram (FPFH) and Triple Orthogonal Local Depth Images (TOLDI). To evaluate their performance, three perturbations were induced: varying surface patch sizes, spatial displacement, and Gaussian deformation. Registration was performed using the RANSAC algorithm. FPFH outperformed TOLDI for small surface patches and in case of Gaussian deformations in terms of registration accuracy. In contrast, TOLDI showed lower registration errors for patches with spatial displacement. While developing a 3D-3D registration pipeline, the choice of the feature descriptor is of importance, consequently a careful choice suitable for the application in LLS is necessary.

Optical Coherence Tomography Angiography (OCTA) is an imaging modality that provides threedimensional information of the retinal microvasculature and therefore promises early diagnosis and sufficient monitoring in ophthalmology. However, there is considerable variability between experts analysing this data. Measures for quantitative assessment of the vasculature need to be developed and established, such as fractal dimension. Fractal dimension can be used to assess the complexity of vessels and has been shown to be independently associated with neovascularization, a symptom of diseases such as diabetic retinopathy. This investigation assessed the performance of three fractal dimension algorithms: Box Counting Dimension (BCD), Information Dimension (ID), and Differential Box Counting (DBC). Two of those, BCD and ID, rely on previous vessel segmentation. Assessment of the added value or disturbance regarding the segmentation step is a second aim of this study. The investigation was performed on a data set composed of 9 in vivo human eyes. Since there is no ground truth available, the performance of the methods in differentiating the Superficial Vascular Complex (SVC) and Deep Vascular Complex (DVC) layers apart and the consistency of measurements of the same layer at different time-points were tested. The performance parameters were the ICC and the Mann- Whitney U tests. The three applied methods were suitable to tell the different layers apart and showed consistent values applied in the same slab. Within the consistency test, the non-segmentation-based method, DBC, was found to be less accurate, expressed in a lower ICC value, compared to its segmentation-based counterparts. This result is thought to be due to the DBC’s higher sensitivity when compared to the other methods. This higher sensitivity might help detect changes in the microvasculature, like neovascularization, but is also more likely prone to noise and artefacts.

Intraoperative blood flow measurement is an effective way to assess the quality of bypass surgery. Flow quantification from indocyanine green (ICG) angiography promises to be an easy, contact-free method. It shows deviations compared to a reference. These are given as factor k, which depends on the vessel diameter d. The radiation transport within the vessel while recording the ICG passage might cause this. It is analyzed in silico to disclose its impact on k (d). A Fluorescence Monte Carlo Multi Cylinder (FMCMC) model was developed as a static model, assuming homogeneous concentration of ICG. In contrast to published approaches utilizing a Monte Carlo Multi Layer (MCML) model assuming the deepest penetration location within a photon packet’s path to be the fluorescence location, the events are modeled. Fluorescence event modeling, Multi Cylinder geometry and a homogeneous illumination as well as combinations of these were implemented in separate aspect models. Resulting k (d) were compared to k (d) from MCML. Deviations in k (d) derived from FMCMC and MCML in each aspect model were present. The Root Mean Square Error ranges from 6,8% to 36 %, k (d) also varied comparing the aspect models to each other. The model geometry, the modeled fluorescence location and illumination mode show a clear impact on simulated k (d). Therefore, our study shows that simplifications of previous studies are invalid. The developed FMCMC model considers the named aspects, allowing the analysis of radiation transport in ICG angiography. The FMCMC model assumes a homogeneous concentration of ICG which is not true in clinical cases. Obtaining the heterogeneous distribution of ICG is possible via fluid flow models. Coupling the fluid flow model and the developed radiation transport model as well as including a detailed camera optic is the task for future work.

This contribution is part of a project concerning the creation of an artificial dataset comprising 3D head scans of craniosynostosis patients for a deep-learning-based classification. To conform to real data, both head and neck are required in the 3D scans. However, during patient recording, the neck is often covered by medical staff. Simply pasting an arbitrary neck leaves large gaps in the 3D mesh. We therefore use a publicly available statistical shape model (SSM) for neck reconstruction. However, most SSMs of the head are constructed using healthy subjects, so the full head reconstruction loses the craniosynostosis-specific head shape. We propose a method to recover the neck while keeping the pathological head shape intact. We propose a Laplace- Beltrami-based refinement step to deform the posterior mean shape of the full head model towards the pathological head. The artificial neck is created using the publicly available Liverpool-York-Model. We apply our method to construct artificial necks for head scans of 50 scaphocephaly patients. Our method reduces mean vertex correspondence error by approximately 1.3 mm compared to the ordinary posterior mean shape, preserves the pathological head shape, and creates a continuous transition between neck and head. The presented method showed good results for reconstructing a plausible neck to craniosynostosis patients. Easily generalized it might also be applicable to other pathological shapes.

Optical Coherence Tomography Angiography (OCTA) is a non-invasive imaging technique that enables the visualization of perfused vasculature in vivo. In ophthalmology, it allows the physician to monitor diseases affecting the vascular networks of the retina such as agerelated macular degeneration or diabetic retinopathy. Due to the complexity of the vasculature in the retina, it is of interest to automatically extract vascular parameters which describe the condition of the vessels. Suitable parameters could improve the diagnosis and the treatment during the course of therapy. We present an automated algorithm to compute the diameters of the vessels in en face OCTA images. After segmenting the images, the vessel centerline was computed using a thinning algorithm. The centerline was refined by detecting invalid pixels such as spurs and by continuing the centerline until the ends of the vessels. Lastly, the diameter was computed by dilating a discrete circle at the position of the centerline or by measuring the distance between both borders of the vessels. The developed algorithms were applied to in vivo images of human eyes. Certainly, no ground truth was available. Hence, a plausibility check was performed by comparing the measured diameters of two different layers of the retina (Superficial Vascular Complex (SVC) and Deep Vascular Complex (DVC)). Each layer exhibits a different characteristic vasculature. The algorithm clearly reflected the differences from both retinal layers. The measured diameters demonstrate that the DVC consists of more capillaries and considerably smaller vessels compared to the SVC. The presented method enables automated analysis of the retinal vasculature and forms thereby the basis for monitoring diseases influencing the vasculature of the retina. The validation of the method using an artificial ground truth is still needed.

Transit times of a bolus through an organ can provide valuable information for researchers, technicians and clinicians. Therefore, an indicator is injected and the temporal propagation is monitored at two distinct locations. The transit time extracted from two indicator dilution curves can be used to calculate for example blood flow and thus provide the surgeon with important diagnostic information. However, the performance of methods to determine the transit time Δt cannot be assessed quantitatively due to the lack of a sufficient and trustworthy ground truth derived from in vivo measurements. Therefore, we propose a method to obtain an in silico generated dataset of differently subsampled indicator dilution curves with a ground truth of the transit time. This method allows variations on shape, sampling rate and noise while being accurate and easily configurable. COMSOL Multiphysics is used to simulate a laminar flow through a pipe containing blood analogue. The indicator is modelled as a rectangular function of concentration in a segment of the pipe. Afterwards, a flow is applied and the rectangular function will be diluted. Shape varying dilution curves are obtained by discrete-time measurement of the average dye concentration over different cross-sectional areas of the pipe. One dataset is obtained by duplicating one curve followed by subsampling, delaying and applying noise. Multiple indicator dilution curves were simulated, which are qualitatively matching in vivo measurements. The curves temporal resolution, delay and noise level can be chosen according to the requirements of the field of research. Various datasets, each containing two corresponding dilution curves with an existing ground truth transit time, are now available. With additional knowledge or assumptions regarding the detection-specific transfer function, realistic signal characteristics can be simulated. The accuracy of methods for the assessment of Δt can now be quantitatively compared and their sensitivity to noise evaluated.

Optical image processing is part of many applications used for brain surgeries. Microscope camera, or patient movement, like brain-movement through the pulse or a change in the liquor, can cause the image processing to fail. One option to compensate movement is feature detection and spatial allocation. This allocation is based on image features. The frame wise matched features are used to calculate the transformation matrix. The goal of this project was to evaluate different feature detectors based on spatial density and temporal robustness to reveal the most appropriate feature. The feature detectors included corner-, and blob-detectors and were applied on nine videos. These videos were taken during brain surgery with surgical microscopes and include the RGB channels. The evaluation showed that each detector detected up to 10 features for nine frames. The feature detector KAZE resulted in being the best feature detector in both density and robustness.

The segmentation and registration of structures are gaining importance due to the increasing demand of automated image enhancement and understanding. Especially in medicine and life science, assistance systems could have a large impact on diagnosis, treatment and quality control. Dye driven procedures, such as fluorescence imaging with Indocyanine green (ICG), are nowadays indispensable because they enhance contrast, reveal structures and deliver the operator with important information. The contact free ICG angiography is providing the surgeon spatial and temporal information on blood flow within a vessel. The processing of those information is done manually or semi automated but is very helpful for the surgeon. Extending the degree of automatism, the amount of information processed and even augment or transfer it into another domain could deliver the operator useful support and improve surgical work flow. Using, analyzing and transferring those information from ICG-IR domain into the RGB domain is the focus of this project. We are introducing a vessel registration method in the RGB domain driven by the spatial fluorescence behavior of the vessel in the ICG-IR domain. The method includes Superpixel based segmentation of the vessel in the ICG-IR domain, the spatial gradient based transfer and registration in the RGB domain and the continuous segmentation of the vessel in a RGB video. This paper show a proof of concept of the method. The results show an successful inter domain information transfer and registration of the vessel. Further tracking of the vessel over all frames is possible. Nevertheless limitations are revealed and discussed.

Image segmentation plays an increasingly important role in image processing. It allows for various applications including the analysis of an image for automatic image understanding and the integration of complementary data. During vascular surgeries, the blood flow in the vessels has to be checked constantly, which could be facilitated by a segmentation of the affected vessels. The segmentation of medical images is still done manually, which depends on the surgeon’s experience and is time-consuming. As a result, there is a growing need for automatic image segmentation methods. We propose an unsupervised method to detect the regions of no interest (RONI) in intraoperative images with low depth-of-field (DOF). The proposed method is divided into three steps. First, a color segmentation using a clustering algorithm is performed. In a second step, we assume that the regions of interest (ROI) are in focus whereas the RONI are unfocused. This allows us to segment the image using an edge-based focus measure. Finally, we combine the focused edges with the color RONI to determine the final segmentation result. When tested on different intraoperative images of aneurysm clipping surgeries, the algorithm is able to segment most of the RONI not belonging to the pulsating vessel of interest. Surgical instruments like the metallic clips can also be excluded. Although the image data for the validation of the proposed method is limited to one intraoperative video, a proof of concept is demonstrated.

Contemporary surgical microscope systems have excellent optical properties but some desirable features remain unavailable. The number of co-observers is currently restricted, by spatial and optical limitations, to only two. Moreover, ergonomics poses are a problem: Current microscope systems impede free movement and sometimes demand that surgeons take uncomfortable postures over long periods of time. To rectify these issues, some companies developed surgical microscope systems based on a streaming approach. These systems remove some of the limitations. Multi-observer positions, for example, are not independent from each other, for example. In order to overcome the aforementioned limitations, we are currently developing an approach for the next generation of surgical microscope: Namely the fully digital surgical microscope, where the current observation system is replaced with a camera array, allowing real-time 3D reconstruction of surgical scenes and, consequently, the rendering of almost unlimited views for multiple observers. These digital microscopes could make the perspective through the microscope unnecessary allowing the surgeon to move freely and work in more comfortable postures. The requirements on the camera array in such a system have to be determined. For this purpose, we propose of estimation the minimal number of cameras and their positions needed for the 3D reconstruction of microsurgical scenes. The method of estimation is based on the requirements for the 3D reconstruction. Within the MATLAB simulation environment, we have developed a 3D model of a microsurgical scene, used for the determination of the number of required cameras. In a next step a small, compact and costefficient system with few opto-mechanical components could be manufactured.

Purpose: Depth estimation is the basis of 3D reconstruction of airway structure from 2D bronchoscopic scenes, which can be further used to develop a vision-based bronchoscopic navigation system. This work aims to improve the performance of depth estimation directly from bronchoscopic images by training a depth estimation network on both synthetic and real datasets. Methods: We propose a cGAN-based network Bronchoscopic-Depth-GAN (BronchoDep-GAN) to estimate depth from bronchoscopic images by translating bronchoscopic images into depth maps. The network is trained in a supervised way learning from synthetic textured bronchoscopic image-depth pairs and virtual bronchoscopic image-depth pairs, and simultaneously, also in an unsupervised way learning from unpaired real bronchoscopic images and depth maps to adapt the model to real bronchoscopic scenes. Results: Our method is tested on both synthetic data and real data. However, the tests on real data are only qualitative, as no ground truth is available. The results show that our network obtains better accuracy in all cases in estimating depth from bronchoscopic images compared to the well-known cGANs pix2pix. Conclusions: Including virtual and real bronchoscopic images in the training phase of the depth estimation networks can improve depth estimation's performance on both synthetic and real scenes. Further validation of this work is planned on 3D clinical phantoms. Based on the depth estimation results obtained in this work, the accuracy of locating bronchoscopes with corresponding pre-operative CTs will also be evaluated in comparison with the current clinical status. [ABSTRACT FROM AUTHOR]