Laboratory for Image & Signal Processing (LISP)
Director: Prof. Dr. Karl Hans Englmeier
Modern medical imaging comes with unique challenges in terms of the identification of objects and structures, image quantification and overall image processing. There is a large diversity of imaging technologies, spanning from non-invasive high-resolution imaging methods such as microscopy, high-resolution intra-operative fluorescence imaging, X-ray Computed Tomography (XCT) and Magnetic Resonance Imaging (MRI) and Photo-Acoustic Imaging (PAI) to low resolution but high specificity modalities such as Positron Emission Tomography (PET), Single Photon Emission Computed Tomography (SPECT) and Fluorescence Molecular Tomography (FMT). Similarly there is an increasing need for novel and fast methods to process huge amounts of data sets produced by these techniques, occasionally in real-time. In addition, novel trends in combining several of these modalities together in multi-modality systems have increased the need for image co-registration and atlas mapping.
The Laboratory for Image & Signal Processing (LISP) focuses on areas of image quantification, segmentation, structure identification, real-time imaging, image co-registration and accelerating processing time. The laboratory plays a key role in the Institute for Biological and Medical Imaging by providing the tools for accurate image handling of existing imaging methods and novel modalities developed, and is integrated in the work flow of providing accurate tools for image segmentation, quantification and visualization for biological and medical imaging needs. In addition the laboratory develops novel processing methods that can more generally be applied to the fields of microscopy, surgical imaging, diagnostic imaging and treatment evaluation, including conventional radiological imaging and molecular imaging. Examples of research directions include:
Model-based Image Segmentation
Image segmentation is an important step in image processing as it allows the identification of structures for enabling automated quantification and classification which are crucial for subsequent processing and visualization. The laboratory develops several automated tools for improving and accelerating segmentation in image processing, largely utilizing model-based methods, as they have proved to be highly successful in the segmentation of even noisy or imperfect images. Typically, a-priori knowledge on the appearance of the structure of interest is incorporated into the segmentation process. This knowledge is gained by analyzing a set of training data which represents all the possible appearances of the specific anatomic structure. Thus, a statistical model can be created which is utilized in the segmentation of unknown image data. Such tools have been applied to the segmentation of organs in X-ray CT and MR images. Representative results are shown in the following.
Fig. 1 Segmentation of µCT mouse images. Left image: Section of a µCT image of a mouse showing femur and knee joint. Right image: 3D model of segmented and labelled bone structures. Red: Patella; light blue: Femur; yellow: Tibia; green: Sesamoid Bones; blue: Minisci. A particular focus here is the computational acceleration, using appropriately developed algorithms, while maintaining a robust imaging platform.
Fig 2: Segmentation of human knee cartilage in magnetic resonance images. The accurate segmentation of this anatomical structure is an important step within clinical studies and drug trials and is crucial for the early diagnosis of conditions like ostheoarthritis. Upper left image: Visualization of MR volume data of a human knee. Upper right image: 3D Model of segmented cartilage tissue. Lower images: Sagittal, axial and coronal slices through the 3D MRI.
Multimodality Image Co-registration
Multiple imaging modalities exist in the clinic today, each offering its own unique advantages to the physician. To be able to take advantage of each imaging systems strength is often challenging to the clinician due to the lack of registration marks native in the scan process or that reside in the patient. Current trials to combine modalities often result in 'home-brewed' systems that are clunky and don't make it past the research/testing phase. To date software methods exist that offer multi-modality registration automatically or with a minimal amount of user intervention but often these still have to be examined with scrutiny, especially in critical area such as head, neck and spinal cord regions.
Furthermore, several novel imaging modalities, currently developed in the IBMI, require new co-registration approaches as well. For instance, photoacoutic imaging is a relatively new technology in the field of biological and medical imaging that can offer functional and molecular contrast mechanisms not present in existing imaging methods. In order to fully benefit from these new imaging technologies and their interplay with other methods, there is a need for efficient algorithms for image reconstruction and registration with data from other imaging modalities like FMT and CT.
Autofocusing in computed microscopy
Microscopy is without doubt the most essential imaging instrument for biologists and is getting more and more important as a tool to achieve research advances in proteomics, genomics, biochemistry and molecular biology. The knowledge gained from these endeavours is crucial for improvements in medical diagnostics, drug discovery and biomedical research. At the same time, handling more samples in quantitative, high-throughput imaging in even shorter time requires Automated Microscopy as manual examination is too time consuming.
Hereby, the Laboratory especially concentrates on developing new Autofocus methods as it is indispensable for the whole microscopy process and its accuracy determines the quality of the overall output. The Laboratory’s efforts aim to overcome the drawbacks of current methods that mainly take the image contrast such as pixel correlation as basis for focus measurements. These approaches work sufficient under well-defined circumstances but lack robustness when applied to more practical and general tasks.
ECG examinations in population-based studies
We performed electrocardiographic (ECG) examinations in population-based surveys in the KORA study area of Augsburg. The data base is comprising more than 20,000 ECGs consisting of 12 lead resting ECG in digital format, quantitative and qualitative computerized ECG analyses and cardiologic data. It also comprises about 5000 rhythm ECG records for heart rate variability analyses. The examinations (some of them under the auspices of the World Health Organisation) were performed in cooperation with the Institute of Epidemiology, the Institute of Human Genetics, the Institute of Health Economy, clinical partners and universities. The computer-processed ECG data base in context with the epidemiologic data, e.g. the ongoing morbidity and mortality follow-up in the study area of Augsburg have been and still are used as a tool for research on
- evaluation of ECG processing methods
- cardiovascular epidemiology
- genetics of left ventricular hypertrophy and rhythm disturbances
- environmental health effects caused by air pollution
- heart rate variability, diabetes and the metabolic syndrome