Sándor Fridli

Radiomics is an emerging field of CT image processing, that offers noninvasive quantification of tumour phenotypes using quantitative image features. Radiomics analysis has promising applications in cancer treatment and personalized medicine, like treatment planning and the prediction of clinical factors. However, the optimal feature selection is not established in the literature, and the applications usually involve data mining of a large pool of features. In this paper, we propose to extract higher-level radiomic features using Radial Harmonic Fourier moments (RHFM). Image moments, and specially orthogonal Fourier moments are widely used in image processing, providing efficient and invariant shape descriptors. In particular, RHFMs are known to perform well on small noisy images, making them a promising candidate for CT tumour analysis. Motivated by these advantages, we developed a feature extraction scheme based on RHFM, and we performed radiomics analysis on lung CT images of non-small cell lung cancer patients. The proposed method is validated on multiple annotated datasets following the literature guidelines, evaluating the accuracy, stability, reliability, and prognostic value of the proposed features. The results show better reliability and otherwise comparable performance compared to the state-of-the-art wavelet descriptors. Furthermore, Fourier moments provide higher level of flexibility and possible adaptivity compared to wavelets, and unlike wavelet features, RHFM features are invariant of position, size and orientation in the tumor region.

We propose a parameterized mathematical model to approximate CT images in the presence of beam hardening. This physical phenomenon causes so-called cupping and streaking artifacts to appear on the reconstructed image. We pose the problem of optimizing the model parameters as a separable nonlinear least squares approximation task. The optimized model parameters have exact physical meaning. More precisely, the model parameters can be used to describe the attenuation properties of different materials with respect to monochromatic components of polychromatic X-ray beams. In addition, the proposed model can be used to correct the reconstructed image in a physically informative way. We find that the proposed method provides reliable estimates of the examined object's attenuation properties up to 4 dominant beam components. We verify the proposed method's effectiveness through numerical simulations.

Sleep is a physiological phenomenon and a sufficient amount of sleep is mandatory for a human for his/her health. The aim of the paper is to develop an efficient algorithm to detect the various sleep states by combining different biomedical signals. The novelty of the algorithm is that we used two piecewise linear data reduction techniques namely a new Halfwave method in time domain and Franklin transformation in frequency domain. The obtained two piecewise linear forms of signals have low complexity that still preserve the characteristics of the stages of the sleep in the signals. The components of the feature vector are generated from the parameters of the two reduced piece wise linear functions. Algorithm is tested on MIT-BIH Polysomnographic Database having more than 70 hours long term EEG, blood and nasal signals with six different sleep classes. Proposed method shows better performance in terms of average Sensitivity, Specificity, Accuracy and False Alarm Rate/hour than state of the art methods.

In this paper we develop an adaptive transform-domain technique based on rational function systems. It is of general importance in several areas of signal theory, including filter design, transfer function approximation, system identification, control theory etc. The construction of the proposed method is discussed in the framework of a general mathematical model called variable projection. First we generalize this method by adding dimension type free parameters. Then we deal with the optimization problem of the free parameters. To this order, based on the well-known particle swarm optimization (PSO) algorithm, we develop the multi-dimensional hyperbolic PSO algorithm. It is designed especially for the rational transforms in question. As a result, the system along with its dimension is dynamically optimized during the process. The main motivation was to increase the adaptivity while keeping the computational complexity manageable. We note that the proposed method is of general nature. As a case study the problem of electrocardiogram (ECG) signal compression is discussed. By means of comparison tests performed on the PhysioNet MIT-BIH Arrhythmia database we demonstrate that our method outperforms other transformation techniques.