Kees Slump

Process identification through modular neural networks and rule extraction

Monolithic neural networks may be trained from measured data to establish knowledge about the process. Unfortunately, this knowledge is not guaranteed to be found and - if at all - hard to extract. Modular neural networks are better suited for this purpose. Domain-ordered by topology, rule extraction is performed module by module. This has all the benefits of a divide-and-conquer method and opens the way to structured design. This paper discusses a next step in this direction by illustrating the potential of base functions to design the neural model.

The Preoperative CT-Scan Can Help to Predict Postoperative Complications after Pancreatoduodenectomy

After pancreatoduodenectomy, complication rates are up to 40%. To predict the risk of developing postoperative pancreatic fistula or severe complications, various factors were evaluated. 110 consecutive patients undergoing pancreatoduodenectomy at our institute between January 2012 and September 2014 with complete CT scan were retrospectively identified. Pre-, per-, and postoperative patients and pathological information were gathered. The CT-scans were analysed for the diameter of the pancreatic duct, attenuation of the pancreas, and the visceral fat area. All data was statistically analysed for predicting POPF and severe complications by univariate and multivariate logistic regression analyses. The POPF rate was 18%. The VFA measured at umbilicus (OR 1.01; 95% CI = 1.00–1.02; P = 0.011) was an independent predictor for POPF. The severe complications rate was 33%. Independent predictors were BMI (OR 1.24; 95% CI = 1.10–1.42; P = 0.001), ASA class III (OR 17.10; 95% CI = 1.60–182.88; P = 0.019), and mean HU (OR 0.98; 95% CI = 0.96–1.00; P = 0.024). In conclusion, VFA measured at the umbilicus seems to be the best predictor for POPF. BMI, ASA III, and the mean HU of the pancreatic body are independent predictors for severe complications following PD.

Simultaneous imaging of ultrasound attenuation, speed of sound, and optical absorption in a photoacoustic setup

Photoacoustic imaging is a relatively new medical imaging modality. In principle it can be used to image the optical absorption distribution of an object by measurements of optically induced acoustic signals. Recently we have developed a modified photoacoustic measurement system which can be used to simultaneously image the ultrasound propagation parameters as well. By proper placement of a passive element we obtain isolated measurements of the objects ultrasound propagation parameters, independent of the optical absorption inside the object. This passive element acts as a photoacoustic source and measurements are obtained by allowing the generated ultrasound signal to propagate through the object. Images of the ultrasound propagation parameters, being the attenuation and speed of sound, can then be reconstructed by inversion of a measurement model. This measurement model relates the projections non-linearly to the unknown images, due to ray refraction effects. After estimating the speed of sound and attenuation distribution, the optical absorption distribution is reconstructed. In this reconstruction problem we take into account the previously estimated speed of sound distribution. So far, the reconstruction algorithms have been tested using computer simulations. The method has been compared with existing algorithms and good results have been obtained.

On the analysis of neural networks for image processing

An integral component of Fuzzy ARTMAPs training phase is the use of Match Tracking (MT), whose functionality is to search for an appropriate category that will correctly classify a presented training pattern in case this particular pattern was originally misclassified. In this paper we explain the MTs role in detail, why it actually works and finally we put its usefulness to the test by comparing it to the simpler, faster alternative of not using MT at all during training. Finally, we present a series of experimental results that eventually raise questions about the MTs utility. More specifically, we show that in the absence of MT the resulting, trained FAM networks are of reasonable size and exhibit better generalization performance.

A maximum likelihood method for obtaining integrated attenuation from ultrasound transmission mode measurements

In photoacoustic imaging applications an image is reconstructed of the optical absorption distribution the imaged object. The photoacoustic measurements however are also dependent on acoustic properties of the imaged object. By estimating the acoustic properties from photoacoustic measurements, we can improve the performance of the optical absorption reconstruction and allow for the imaging of separate acoustic modalities. We derive and evaluate a maximum likelihood estimator for the measurements of integrated acoustic attenuation. This estimator is applicable to media like soft tissue. In this kind of media, the attenuation due to dissipative effects obeys a frequency power law. By measuring the propagation of ultrasound signals through such a medium, the parameters that describe the attenuation can be estimated. In this paper a new method is introduced for estimating the attenuation of ultrasound media by means of transmission mode measurements. The method is based on analyzing the noise characteristics...

Public safety and cognitive radio

This book gives comprehensive and balanced coverage of the principles of cognitive radio communications, cognitive networks, and details of their implementation, including the latest developments in the standards and spectrum policy. Case studies, end-of-chapter questions, and descriptions of various platforms and test beds, together with sample code, give hands-on knowledge of how cognitive radio systems can be implemented in practice. Extensive treatment is given to several standards, including IEEE 802.22 for TV White Spaces and IEEE SCC41.