Within MedLab - and especially the Therapy Planning group at ZIB - we are dealing with a variety of medical and anthropometric data. To tackle the challenges of analyzing an always increasing amount of data and to provide software tools to automatically extract the relevant information out of it, we are investigating model based approaches (statistical shape and appearance models) as well as machine learning techniques (regression, classification, and semi-supervised learning), which can then be used in a number of applications such as scene recognition from photographs, object recognition in images, parameter estimation, classification, up to automatic diagnosis from medical image data.
The project is in close collaboration with 1000shapes GmbH, a ZIB spin-off that transfers research in life sciences into products for clinical applications. Within the project, algorithms are to be identified, developed/implemented, compared, and tested on data from various application domains. The successful applicant will have the opportunity to perform research in medical image computing and geometry processing within the ZIB research group Therapy Planning while obtaining professional support from 1000shapes in software development with insights into relevant applications. Within the project, students will have the opportunity to experience medical research under professional supervision in combination with industry-strength software development, with the goal of practical solutions and publication.
Building on large medical image as well as a anthropometric 3D face databases, students will have the opportunity to investigate machine-learning approaches, i.e. deep learning, or the application of regression forests, to identify, analyze, and classify features or patterns based on medical images or geometric models.
Several databases are the foundation for possible investigations: (1) The OAI database of the Osteoarthritis Initiative (OAI), which is a multi-center, longitudinal, prospective observational study of knee osteoarthritis, providing clinical evaluation data, radiological (x-ray and magnetic resonance) images, and a bio-specimen repository from over 5000 patients. This information has great potential, both for developing a better understanding of disease onset and progression, as well as improving future therapeutic concepts; (2) A database of several hundreds of 3D face models from various individuals and with varying facial expressions, providing information for anthropometrical studies or psychological experiments; (3) A huge collection of dental 3D image data, where bony structures, nerves, and teeth are to be segmented, anatomical relationships are to be analyzed, and suitable shape and appearance models are to be developed; (4) A database of human spines giving the opportunity to study the morphology of single vertebrae up to the complete spine as well as the functional performance within the context of biomechanics and orthopedic research. The processing of such databases requires automated image and geometry processing as well as sophisticated data analysis approaches.
Challenges: For medical image processing, the challenge is to automatically extract anatomical shape and appearance information from image data, as well as to integrate this information in so called statistical 3D shape and appearance models to train and improve automated algorithms. For geometric data the challenge is to improve methods for determining correspondences, to analyze variation in shape, to establish suitable metrics for measuring similarities in various shape spaces, for clustering and population based analysis. To this end, machine learning combined with model-based approaches shall be employed and adopted.
Example project: Establishing correspondences between shapes lies at the core of many operations in image analysis and geometry processing. The majority of existing methods formulate the matching problem as finding optimal pairings of points or regions on shapes. This representation, however, renders the matching intractable as the space of possible point correspondences grows exponentially and does not naturally support constraints such as map continuity or global consistency. Within a possible project, we will investigate a recent alternative approach that generalizes the notion of correspondences to mappings between real-valued functions on the shapes rather than the standard point-to-point maps. One challenge that we will address is the adaptation of this functional maps framework to the matching of volumetric geometries from 3D medical image data. Based on this, we will further derive a scheme for inferring group-wise correspondences that takes advantage of the context provided by the collection of shapes.
Based on the aforementioned background several topics in medical image and geometry processing are conceivable. Results of our research may become a basis for improving existing segmentation, classification, and diagnosis algorithms that are currently under development by 1000shapes.
The prospective participant should:
- have a background in computer science, bio-engineering, mathematics or physics,
- preferably have experience with collaborative code development in C++ on Windows or Linux,
- have attended classes in the area of machine learning, computer vision, image processing or statistics,
- preferably be acquainted with software for processing large medical image datasets,
- have fun working in teams, both with colleagues from academia and industry,
be able to communicate fluently in English