Despite the advances in automated medical image segmentation, AI models still underperform in various clinical settings, posing challenges for integration into real-world workflows. In this pre-registered prospective multicenter evaluation, we analyzed 20 state-of-the-art mandibular segmentation models across 19,218 segmentations of 1,000 clinically resampled CT/CBCT scans. Our results suggest that for a given model, segmentation accuracy can vary by up to 25% in Dice score as socio-technical factors such as voxel size, bone orientation, and patient conditions (e.g., osteosynthesis or pathology) shift from favorable to adverse. Higher sharpness, isotropic smaller voxels, and neutral orientation significantly improved results, while metallic osteosynthesis and anatomical complexity led to significant degradation. Our findings challenge the common view of AI models as “plug-and-play” tools and suggest evidence-based optimization recommendations for both clinicians and developers. This will in turn boost the integration of AI segmentation tools in routine healthcare.

Image-based patient-specific modelling of hemodynamics are gaining increased popularity as a diagnosis and outcome prediction solution for a variety of cardiovascular diseases. While their potential to improve diagnostic capabilities and thereby clinical outcome is widely recognized, these methods require considerable computational resources since they are mostly based on conventional numerical methods such as computational fluid dynamics (CFD). As an alternative to the numerical methods, we propose a machine learning (ML) based approach to calculate patient-specific hemodynamic parameters. Compared to CFD based methods, our approach holds the benefit of being able to calculate a patient-specific hemodynamic outcome instantly with little need for computational power. In this proof-of-concept study, we present a deep artificial neural network (ANN) capable of computing hemodynamics for patients with aortic coarctation in a centerline aggregated (i.e., locally averaged) form. Considering the complex relation between vessels shape and hemodynamics on the one hand and the limited availability of suitable clinical data on the other, a sufficient accuracy of the ANN may however not be achieved with available data only. Another key aspect of this study is therefore the successful augmentation of available clinical data. Using a statistical shape model, additional training data was generated which substantially increased the ANN's accuracy, showcasing the ability of ML based methods to perform in-silico modelling tasks previously requiring resource intensive CFD simulations.

The aim of this paper is to provide a comprehensive overview of the MICCAI 2020 AutoImplant Challenge. The approaches and publications submitted and accepted within the challenge will be summarized and reported, highlighting common algorithmic trends and algorithmic diversity. Furthermore, the evaluation results will be presented, compared and discussed in regard to the challenge aim: seeking for low cost, fast and fully automated solutions for cranial implant design. Based on feedback from collaborating neurosurgeons, this paper concludes by stating open issues and post-challenge requirements for intra-operative use. The codes can be found at https://github.com/Jianningli/tmi.

Modeling of hemodynamics and artificial intelligence have great potential to support clinical diagnosis and decision making. While hemodynamics modeling is extremely time- and resource-consuming, machine learning (ML) typically requires large training data that are often unavailable. The aim of this study was to develop and evaluate a novel methodology generating a large database of synthetic cases with characteristics similar to clinical cohorts of patients with coarctation of the aorta (CoA), a congenital heart disease associated with abnormal hemodynamics. Synthetic data allows use of ML approaches to investigate aortic morphometric pathology and its influence on hemodynamics. Magnetic resonance imaging data (154 patients as well as of healthy subjects) of aortic shape and flow were used to statistically characterize the clinical cohort. The methodology generating the synthetic cohort combined statistical shape modeling of aortic morphometry and aorta inlet flow fields and numerical flow simulations. Hierarchical clustering and non-linear regression analysis were successfully used to investigate the relationship between morphometry and hemodynamics and to demonstrate credibility of the synthetic cohort by comparison with a clinical cohort. A database of 2652 synthetic cases with realistic shape and hemodynamic properties was generated. Three shape clusters and respective differences in hemodynamics were identified. The novel model predicts the CoA pressure gradient with a root mean square error of 4.6 mmHg. In conclusion, synthetic data for anatomy and hemodynamics is a suitable means to address the lack of large datasets and provide a powerful basis for ML to gain new insights into cardiovascular diseases.

We propose a novel GPU-based approach to render virtual X-ray projections of deformable tetrahedral meshes. These meshes represent the shape and the internal density distribution of a particular anatomical structure and are derived from statistical shape and intensity models (SSIMs). We apply our method to improve the geometric reconstruction of 3D anatomy (e.g. pelvic bone) from 2D X-ray images. For that purpose, shape and density of a tetrahedral mesh are varied and virtual X-ray projections are generated within an optimization process until the similarity between the computed virtual X-ray and the respective anatomy depicted in a given clinical X-ray is maximized. The OpenGL implementation presented in this work deforms and projects tetrahedral meshes of high resolution (200.000+ tetrahedra) at interactive rates. It generates virtual X-rays that accurately depict the density distribution of an anatomy of interest. Compared to existing methods that accumulate X-ray attenuation in deformable meshes, our novel approach significantly boosts the deformation/projection performance. The proposed projection algorithm scales better with respect to mesh resolution and complexity of the density distribution, and the combined deformation and projection on the GPU scales better with respect to the number of deformation parameters. The gain in performance allows for a larger number of cycles in the optimization process. Consequently, it reduces the risk of being stuck in a local optimum. We believe that our approach will improve treatments in orthopedics, where 3D anatomical information is essential.

This paper presents a comparison study between 10 automatic and six interactive methods for liver segmentation from contrast-enhanced CT images. It is based on results from the "MICCAI 2007 Grand Challenge" workshop, where 16 teams evaluated their algorithms on a common database. A collection of 20 clinical images with reference segmentations was provided to train and tune algorithms in advance. Participants were also allowed to use additional proprietary training data for that purpose. All teams then had to apply their methods to 10 test datasets and submit the obtained results. Employed algorithms include statistical shape models, atlas registration, level-sets, graph-cuts and rule-based systems. All results were compared to reference segmentations five error measures that highlight different aspects of segmentation accuracy. All measures were combined according to a specific scoring system relating the obtained values to human expert variability. In general, interactive methods reached higher average scores than automatic approaches and featured a better consistency of segmentation quality. However, the best automatic methods (mainly based on statistical shape models with some additional free deformation) could compete well on the majority of test images. The study provides an insight in performance of different segmentation approaches under real-world conditions and highlights achievements and limitations of current image analysis techniques.

Abstract We present an approach for extracting extremal feature lines of scalar indicators on surface meshes, based on discrete Morse Theory. By computing initial Morse‐Smale complexes of the scalar indicators of the mesh, we obtain a candidate set of extremal feature lines of the surface. A hierarchy of Morse‐Smale complexes is computed by prioritizing feature lines according to a novel criterion and applying a cancellation procedure that allows us to select the most significant lines. Given the scalar indicators on the vertices of the mesh, the presented feature line extraction scheme is interpolation free and needs no derivative estimates. The technique is insensitive to noise and depends only on one parameter: the feature significance. We use the technique to extract surface features yielding impressive, non photorealistic images.