Theses from Medical Informatics

Electronic health record systems (EHR) are used to store relevant heath facts about patients. The main use of the EHR is in the care of the patient, but an additional use is to reuse the EHR information to locate and evaluate clinical evidence for treatments. To efficiently use the EHR information it is essential to use appropriate methods for information compilations. This thesis deals with use of information in medical terminology systems and ontologies to be able to better use and reuse EHR information and other medical information.

The first objective of the thesis is to examine if word alignment on bilingual English-Swedish rubrics from five medical terminology systems can be used to build a bilingual dictionary. A study found that it was possible to generate a dictionary with 42 000 entries containing a high proportion of medical entries using word alignment. The method worked best using sets of rubrics with many unique words that are consistently translated. The dictionary can be used as a general medical dictionary, for use in semi-automatic translation methods, for use in cross-language information retrieval systems, and for enrichment of other terminology systems.

The second objective of the thesis is to explore how connections from existing terminology systems and information models to SNOMED CT and the structure in SNOMED CT can be used to reuse information. A study examined whether the primary health care diagnose terminology system KSH97-P can obtain a richer structure using category and chapter mappings from KSH97-P to SNOMED CT and the structure in SNOMED CT. The study showed that KSH97-P can be enriched with a poly-hierarchical chapter division and additional attributes. The richer structure was used to compile statistics in new manners that showed new views of the primary care diagnoses. A literature study evaluated which kinds of information compilations those are necessary to create graphical patient overviews based on information from EHRs. It was found that a third of the patient overviews can have their information needs satisfied using compilations based on SNOMED CT encodings of the information entities in the EHR and the structure in SNOMED CT. The other overviews also need access to individual values in the EHR. This can be achieved by using well-defined information models in the EHR.

Methods for measuring deformation and motion of the human heart in-vivo are crucial in the assessment of cardiac function. Applications ranging from basic physiological research, through early detection of disease to follow-up studies, all rely on the quality of the measurements of heart dynamics. This thesis presents new improved magnetic resonance imaging methods for acquisition, image reconstruction and visualization of cardiac motion and deformation.As the heart moves and changes shape during the acquisition, synchronization to the heart dynamics is necessary. Here, a method to resolve not only the cardiac cycle but also the respiratory cycle is presented. Combined with volumetric imaging, this produces a five-dimensional data set with two cyclic temporal dimensions. This type of data reveals unique physiological information, such as interventricular coupling in the heart in different phases of the respiratory cycle.The acquisition can also be sensitized to motion, measuring not only the magnitude of the magnetization but also a signal proportional to local velocity or displacement. This allows for quantification of the motion which is especially suitable for functional study of the cardiac deformation. In this work, an evaluation of the influence of several factors on the signal-to-noise ratio is presented for in-vivo displacement encoded imaging. Additionally, an extension of the method to acquire multiple displacement encoded slices in a single breath hold is also presented.Magnetic resonance imaging is usually associated with long scan times, and many methods exist to shorten the acquisition time while maintaining acceptable image quality. One class of such methods involves acquiring only a sparse subset of k-space. A special reconstruction is then necessary in order to obtain an artifact-free image. One family of these reconstruction techniques tailored for dynamic imaging is the k-t BLAST approach, which incorporates data-driven prior knowledge to suppress aliasing artifacts that otherwise occur with the sparse sampling. In this work, an extension of the original k-t BLAST method to two temporal dimensions is presented and applied to data acquired with full coverage of the cardio-respiratory cycles. Using this technique, termed k-t² BLAST, simultaneous reduction of scan time and improved spatial resolution is demonstrated. Further, the loss of temporal fidelity when using the k-t BLAST approach is investigated, and an improved reconstruction is proposed for the application of cardiac function analysis.Visualization is a crucial part of the imaging chain. Scalar data, such as regular anatomical images, are straightforward to display. Myocardial strain and strain-rate, however, are tensor quantities which do not lend themselves to direct visualization. The problem of visualizing the tensor field is approached in this work by combining a local visualization that displays all degrees of freedom for a single tensor with an overview visualization using a scalar field representation of the complete tensor field. The scalar field is obtained by iterated adaptive filtering of a noise field, creating a continuous geometrical representation of the myocardial strain-rate tensor field.The results of the work presented in this thesis provide opportunities for improved imaging of myocardial function, in all areas of the imaging chain; acquisition, reconstruction and visualization.

A Riemannian manifold is a mathematical concept that generalizes curved surfaces to higher dimensions, giving a precise meaning to concepts like angle, length, area, volume and curvature. A glimpse of the consequences of a non-flat geometry is given on the sphere, where the shortest path between two points – a geodesic – is along a great circle. Different from Euclidean space, the angle sum of geodesic triangles on the sphere is always larger than 180 degrees.

Signals and data found in applied research are sometimes naturally described by such curved spaces. This dissertation presents basic research and tools for the analysis, processing and visualization of such manifold-valued data, with a particular emphasis on future applications in medical imaging and visualization.

Two-dimensional manifolds, i.e. surfaces, enter naturally into the geometric modelling of anatomical entities, such as the human brain cortex and the colon. In advanced algorithms for processing of images obtained from computed tomography (CT) and ultrasound imaging (US), images themselves and derived local structure tensor fields may be interpreted as two- or three-dimensional manifolds. In diffusion tensor magnetic resonance imaging (DT-MRI), the natural description of diffusion in the human body is a second-order tensor field, which can be related to the metric of a manifold. A final example is the analysis of shape variations of anatomical entities, e.g. the lateral ventricles in the brain, within a population by describing the set of all possible shapes as a manifold.

Work presented in this dissertation include: Probabilistic interpretation of intrinsic and extrinsic means in manifolds. A Bayesian approach to filtering of vector data, removing noise from sampled manifolds and signals. Principles for the storage of tensor field data and learning a natural metric for empirical data.

The main contribution is a novel class of algorithms called LogMaps, for the numerical estimation of logp (x) from empirical data sampled from a low-dimensional manifold or geometric model embedded in Euclidean space. The logp (x) function has been used extensively in the literature for processing data in manifolds, including applications in medical imaging such as shape analysis. However, previous approaches have been limited to manifolds where closed form expressions of logp (x) have been known. The introduction of the LogMap framework allows for a generalization of the previous methods. The application of LogMaps to texture mapping, tensor field visualization, medial locus estimation and exploratory data analysis is also presented.

In medicine, data are produced from different sources and continuously stored in data depositories. Examples of these growing databases are quality registries. In Sweden, there are many cancer registries where data on cancer patients are gathered and recorded and are used mainly for reporting survival analyses to high level health authorities.

In this thesis, a breast cancer quality registry operating in South-East of Sweden is used as the data source for newer analytical techniques, i.e. data mining as a part of knowledge discovery in databases (KDD) methodology. Analyses are done to sift through these data in order to find interesting information and hidden knowledge. KDD consists of multiple steps, starting with gathering data from different sources and preparing them in data pre-processing stages prior to data mining.

Data were cleaned from outliers and noise and missing values were handled. Then a proper subset of the data was chosen by canonical correlation analysis (CCA) in a dimensionality reduction step. This technique was chosen because there were multiple outcomes, and variables had complex relationship to one another.

After data were prepared, they were analyzed with a data mining method. Decision tree induction as a simple and efficient method was used to mine the data. To show the benefits of proper data pre-processing, results from data mining with pre-processing of the data were compared with results from data mining without data pre-processing. The comparison showed that data pre-processing results in a more compact model with a better performance in predicting the recurrence of cancer.

An important part of knowledge discovery in medicine is to increase the involvement of medical experts in the process. This starts with enquiry about current problems in their field, which leads to finding areas where computer support can be helpful. The experts can suggest potentially important variables and should then approve and validate new patterns or knowledge as predictive or descriptive models. If it can be shown that the performance of a model is comparable to domain experts, it is more probable that the model will be used to support physicians in their daily decision-making. In this thesis, we validated the model by comparing predictions done by data mining and those made by domain experts without finding any significant difference between them.

Breast cancer patients who are treated with mastectomy are recommended to receive radiotherapy. This treatment is called postmastectomy radiotherapy (PMRT) and there is a guideline for prescribing it. A history of this treatment is stored in breast cancer registries. We analyzed these datasets using rules from a clinical guideline and identified cases that had not been treated according to the PMRT guideline. Data mining revealed some patterns of non-compliance with the PMRT guideline. Further analysis with data mining revealed some reasons for guideline non-compliance. These patterns were then compared with reasons acquired from manual inspection of patient records. The comparisons showed that patterns resulting from data mining were limited to the stored variables in the registry. A prerequisite for better results is availability of comprehensive datasets.

Medicine can take advantage of KDD methodology in different ways. The main advantage is being able to reuse information and explore hidden knowledge that can be obtained using advanced analysis techniques. The results depend on good collaboration between medical informaticians and domain experts and the availability of high quality data.

We present a novel method for manifold learning, i.e. identification of the low-dimensional manifold-like structure present in a set of data points in a possibly high-dimensional space. The main idea is derived from the concept of Riemannian normal coordinates. This coordinate system is in a way a generalization of Cartesian coordinates in Euclidean space. We translate this idea to a cloud of data points in order to perform dimension reduction. Our implementation currently uses Dijkstra's algorithm for shortest paths in graphs and some basic concepts from differential geometry. We expect this approach to open up new possibilities for analysis of e.g. shape in medical imaging and signal processing of manifold-valued signals, where the coordinate system is “learned” from experimental high-dimensional data rather than defined analytically using e.g. models based on Lie-groups.

We propose a novel post processing method for visualization of fiber traces from DT-MRI data. Using a recently proposed non-linear dimensionality reduction technique, Laplacian eigenmaps (Belkin and Niyogi, 2002), we create a mapping from a set of fiber traces to a low dimensional Euclidean space. Laplacian eigenmaps constructs this mapping so that similar traces are mapped to similar points, given a custom made pairwise similarity measure for fiber traces. We demonstrate that when the low-dimensional space is the RGB color space, this can be used to visualize fiber traces in a way which enhances the perception of fiber bundles and connectivity in the human brain.