BIOP Research Program
Biomedical image and data processing
Imaging and inference, based on signals and data generated from various imaging systems investigated in BIOP, also requires extensive processing. In our cases, signals are often two- or three-dimensional images, but they may also be measurements taken in points represented as spectra. Nevertheless, the common factor for these systems is that the amount of data generated is often quite large, and therefore data processing is inevitable. For example, analysis of heartbeat signals is required to estimate the degree of calcification of the coronary artery. Another example is application of image processing and noise reduction techniques in mammography for improved diagnosis of breast cancer.
Furthermore, image processing may be applied to enhance certain features in these images. The latter implies the development of automatic tools operating on preprocessed data. Development of models, which incorporates a priori knowledge, such as size, shape, or relative displacements, is essential for the operation of an automatic system for segmentation and pattern recognition in images. Applying such methods facilitates development of low-level segmentation algorithms including noise filtering. Moreover, the segmented objects should be transformed into a so-called feature space in which classification may take place. In this part, close collaboration with clinicians is imperative in order set up proper models.
The classification algorithms used in this work may be either decision trees or discriminant filters. For more complex problems, it may be necessary to apply more complex classification algorithms, such as artifical neural networks, which are nonlinear non-parametric models. These models are often well suited in so-called sensor fusion, which covers the fact that the desired information about the biological system is found using different measurements and signals. An important part of the data processing in biomedical optics is visualization, as for example three-dimensional histology performed with optical coherence tomography systems, which is generally labeled computer-aided diagnosis.
The main purpose of this focus area is to develop new techniques for image and data processing in order to enhance the systems investigated in other focus areas of BIOP.
Our current activities are concentrated on:
- noise-reduction algorithms for image enhancement,
- development of sub-space classification algorithms for biomedical data.