A complete CAD system incorporating novel features is presented, and its performance with three separate classifiers is compared and analyzed. The overall performance of our CAD system equipped with any of the three classifiers is well with respect to other methods described in literature.

No abstract

No abstract

Our objective is to automate the detection of apnea and hypopnea events in obstructive sleep apnea hypopnea (OSAH) syndrome based on analysis of arterial oxygen saturation signal alone. This is the first attempt where wavelet is used to detect OSAH events. Detection of OSAH events through wavelet depends on the fluctuations in the magnitude of the transformed coefficients, thus circumventing the problem of variability in the criteria on the magnitude and duration of the signal. Our work evaluates the performance of the wavelet transform to detect OSAH events against three conventional amplitude and duration algorithms. High performance in the detection of OSAH events can be achieved through the wavelet algorithm (score 96.55%, sensitivity 95.74% and specificity 97.02%) if the threshold on wavelet coefficients is individually tuned for each study. However, this is impossible in clinical practice. It is interesting to observe that the conventional methods based on amplitude and duration are able to attain a performance as close as this. The Nervus algorithm obtains the best result (score 96.66%, sensitivity 95.26% and specificity 97.46%) compared to the amplitude duration algorithm, the drop duration algorithm and the wavelet algorithm with global threshold, in descending order of performance.

Feature selection is a task of great importance. Many feature selection methods have been proposed, and can be divided generally into two groups based on their dependence on the learning algorithm/classifier. Recently, a feature selection method that selects features at the same time as it evolves neural networks that use those features as inputs called Feature Selective NeuroEvolution of Augmenting Topologies (FS-NEAT) was proposed by Whiteson et al. In this paper, a novel feature selection method called Feature Deselective NeuroEvolution of Augmenting Topologies (FD-NEAT) is presented. FD-NEAT begins with fully connected inputs in its networks, and drops irrelevant or redundant inputs as evolution progresses. Herein, the performances of FD-NEAT, FS-NEAT and traditional NEAT are compared in some mathematical problems, and in a challenging race car simulator domain (RARS). On the whole, the results show that FD-NEAT significantly outperforms FS-NEAT in terms of network performance and feature selection, and evolves networks that offer the best compromise between network size and performance.

MatLab® has often been considered an excellent environment for fast algorithm development but is generally perceived as slow and hence not fit for routine medical image processing, where large data sets are now available e.g., high-resolution CT image sets with typically hundreds of 512x512 slices. Yet, with proper programming practices – vectorization, pre-allocation and specialization – applications in MatLab® can run as fast as in C language. In this article, this point is illustrated with fast implementations of bilinear interpolation, watershed segmentation and volume rendering.

This paper attempts to improve the diagnostic quality of magnetic resonance (MR) images through application of lossy compression as a noise-reducing filter. The amount of imaging noise present in MR images is compared with the amount of noise introduced by the compression, with particular attention given to the situation where the compression noise is a fraction of the imaging noise. A popular wavelet-based algorithm with good performance, Set Partitioning in Hierarchical Trees (SPIHT), was employed for the lossy compression. Tests were conducted with a number of MR patient images and corresponding phantom images. Different plausible ratios between imaging noise and compression noise (ICR) were considered, and the achievable compression gain through the controlled lossy compression was evaluated. Preliminary results show that at certain ICR's, it becomes virtually impossible to distinguish between the original and compressed-decompressed image. Radiologists presented with a blind test, in certain cases, showed preference to the compressed image rather than the original uncompressed ones, indicating that under controlled circumstances, lossy image compression can be used to improve the diagnostic quality of the MR images.

No abstract