Visual Analytics for Segmentation of 3D Medical Images
The aim of this project is to develop new Visual Analytics methods for fitting statistical shape models (SSMs). These models are helful in various medical applications, e.g., in treatment planning or computer-aided diagnosis. For building an SSM, models have to be selected that fit the training data well. Due to the lack of prior information, standard models are frequently chosen which may not describe the data in an optimal way. We develop new Visual analytics methods for improving this modeling process.
Cooperation: Prof. Georgios Sakas, TU Darmstadt, Medical Imaging Group.
- Extended Surface Distance for Local Evaluation of 3D Medical Image Segmentations
- Visual Comparison of 3D Medical Image Segmentation Algorithms Based on Statistical Shape Models
- Opening up the “black box” of medical image segmentation with statistical shape models
- Visual Analytics for model-based medical image segmentation: Opportunities and challenges
Extended Surface Distance for Local Evaluation of 3D Medical Image Segmentations
The evaluation of 3D medical image segmentation quality requires a reliable detailed comparison of a reference segmentation with an automatic segmentation. The widely used local measure – Surface Distance has significant drawbacks such as asymmetry and underestimation in distant or differently formed regions.
We present Extended Surface Distance, which is a more reliable distance measure for assessing and analyzing local differences between automatic and reference (i.e., ground truth) 3D segmentations.
software, Paper (PDF)
Visual Comparison of 3D Medical Image Segmentation Algorithms Based on Statistical Shape Models
3D medical image segmentation is needed for diagnosis and treatment. For finding best segmentation algorithms, several algorithms need to be evaluated on a set of organ instances.
We present a novel method for comparison and evaluation of several algorithms that automatically segment 3D medical images. It combines algorithmic data analysis with interactive data visualization. A clustering algorithm identi es regions of common quality across the segmented data set for each algorithm. The comparison identi es best algorithms per region. Interactive views show the algorithm quality.
Paper (PrePrint PDF)
Opening up the “black box” of medical image segmentation with statistical shape models
For high quality automatic segmentations, algorithms based on statistical shape models (SSMs) are often used. They segment the image in an iterative way. However, segmentation experts and other users can only asses the final segmentation results, as the segmentation is performed in a “black box manner”.
We present a novel Visual Analytics method, which offers deeper insight into the image segmentation. It allows the expert to assess the quality development (convergence) of the model both on global (full organ) and local (organ areas, landmarks) level. Thereby, important patterns can be found e.g., non-converging parts of the organ during the segmentation. The localization and specifications of such problems helps the experts creating segmentation algorithms to identify algorithm drawbacks and thus it may point out possible ways how to improve the algorithms
video, Paper (PDF)
Visual Analytics for model-based medical image segmentation: Opportunities and challenges
Segmentation of medical images is a prerequisite in clinical practice. Many segmentation algorithms use statistical shape models. Model-based segmentation can be supported by Visual Analytics tools, which give the user a deeper insight into the correspondence between data and model result. Combining both approaches, better models for segmentation of organs in medical images are created.
In this work, we identify the main tasks and problems in model-based image segmentation. As a proof of concept, we show that already small visual-interactive extensions can be very beneficial. Based on these results, we present research challenges for Visual Analytics in this area.