Catálogo de publicaciones - libros

Bias field signal is a low-frequency and very smooth signal that corrupts MRI images specially those produced by old MRI () machines. Image processing algorithms such as segmentation, texture analysis or classification that use the graylevel values of image pixels will not produce satisfactory results. A pre-processing step is needed to correct for the bias field signal before submitting corrupted MRI images to such algorithms or the algorithms should be modified. In this report we discuss two approaches to deal with bias field corruption. The first approach can be used as a preprocessing step where the corrupted MRI image is restored by dividing it by an estimated bias field signal using a surface fitting approach. The second approach shows how to modify the fuzzy -means algorithm so that it can be used to segment an MRI image corrupted by a bias field signal.

We propose a new very simple method to determine roughness of a surface of an implant material from its scanning electron microscopy (SEM) image. For this purpose we have combined a preprocessing method that has been used in histopathology with fractal method used in nonlinear time series analysis. In the pre-processing step the image is transformed into 1-D signals (‘landscapes’) that are subsequently analyzed. Our method draws from multiple disciplines and may find multidisciplinary applications.

The paper presents a novel approach to the problem of reliable estimation of the output flow going out from pneumatically controlled ventricular assist device (VAD). Among many possibilities, the application of artificial neural network (ANN) has been decided leading to the promising results. The basic difficulty however is to perform the suitable sufficiently exact measurement on the pneumatic side of the assisting system, which allows to avoid the application of the e.g. ultrasound measuring transducers on the hydraulic side, and which makes possible an implementation of the automatic control algorithm in the future for the whole measurement process. It is important however to underline that from the physical properties point of view these two mentioned sides i.e. pneumatic and hydraulic are completely different. Therefore, due to several nonlinearities, application of ANN gives an acceptable solution.

The proper balance between a symmetric and asymmetric cell division is crucial for the neural stem cell maintenance both and . These conditions are provided by specific regions of the brain called neural stem cell niches and occur in neurospheres or adherent clones. A method and a tool for cell culture growth monitoring applied in the investigation of the clonally growth of HUCB-NSC (Human Umbilical Cord Blood derived Neural Stem Cells) line, as an model of the neural stem cell niche, is proposed.

In this paper neural classifier system preliminary feature extraction and selection process using time-frequency representation of heart rate variability (HRV) signal is presented. The crucial point of described method is hybrid multi-domain feature set creation, combining different type parameters as well as feature selection based on the measure of class separability property, computed for each extracted feature. Regarding specific properties of non-stationary HRV signal, wavelet transform was chosen as time-frequency representation tool. Presented results are connected both with optimal feature extraction and selection of HRV signals from patient with coronary artery disease as well as classifier performance verification.

Proper interpretation of results of clustering of gene expression data from DNA microarray tests is one of major challenges in experiment data analysis. Interpretation problems arise due to the fact that different algorithms tend to produce different results, while some clusters appear to be invariant of an algorithm applied. A procedure described in this work can be a good starting point for a decision making process to evaluate biological relevance of clustering results obtained. In our view, any other similar approach aiming to discover biologically relevant clusters will have to include biological information. It would be probably beneficial if relevant biological knowledge could be incorporated on the input side of clustering algorithm rather than at the results post processing / interpretation stage, as described in this work. Making clustering algorithms make clustering decision biased towards biologically relevant groupings, thus forming ‘supervised clustering’ approach may be a motivation for further research in this area.

A computer system has been developed for evaluating the morphometrical feature extraction. The features are derived directly from a digital scan of breast fine needle biopsy slides. First the background elimination by thresholding hue component is applied, then the actual segmentation is done with region growing technique. The quality of feature space is measured with classifier based on nonparametric density estimation. The automatic system of malignancy classification was applied on a set of medical images with promising results. The comparison of human accuracy in the cytological diagnosis of breast cancer with the accuracy of digital image analysis combined with computer-based classification is presented.

In this paper, a knowledge discovery framework is used for protein classification. The processing is achieved in three steps: feature extraction, feature ranking and feature selection. Inspirited from text mining results for the first step, we use -grams descriptors; descriptors are ranked from chi-2 statistical indices in the second step; and in the final step, the subset of descriptors is selected which will minimize the prediction error rate using a k-nearest neighbor classifier. Experiments show that this framework gives good results: the dimensionality reduction is effective and increases the classifier performances.

The study presents automatic identification of lungs area and rib borders detection in digital chest radiograph. There are used several segmentation methods, region labeling and region convex hull construction algorithm in lungs identification. Their final shape is derived from lungs identification algorithm, which uses machine learning. Ribs’ borders detection is based on Canny’s edge detection algorithm. Image of chest radiograph processed by Canny’s algorithm is used as entry point to the ribs’ border detection. Then, there are presented successive stages of borders detection for a sample rib. There is also described a method of gathering new patterns of ribs’ borders.

A simple and fast method is presented which detects mitotic cells from two cell lines imaged in two phase-contrast microscopy techniques. Such detection is a first step to more sophisticated image processing tasks like determination of mitotic index or mitotic cell tracking in time-lapse movies. Detection algorithm is based on template matching approach that provides a list of candidates. The list is then pruned by validation algorithm that takes into account information about mitotic cells. The method has been implemented as plugin for ImageJ and has been tested for several different data sets.