Anthony D. Walmsley

The automatic recognition and counting of cough

BackgroundCough recordings have been undertaken for many years but the analysis of cough frequency and the temporal relation to trigger factors have proven problematic. Because cough is episodic, data collection over many hours is required, along with real-time aural analysis which is equally time-consuming.A method has been developed for the automatic recognition and counting of coughs in sound recordings.MethodsThe Hull Automatic Cough Counter (HACC) is a program developed for the analysis of digital audio recordings. HACC uses digital signal processing (DSP) to calculate characteristic spectral coefficients of sound events, which are then classified into cough and non-cough events by the use of a probabilistic neural network (PNN). Parameters such as the total number of coughs and cough frequency as a function of time can be calculated from the results of the audio processing.Thirty three smoking subjects, 20 male and 13 female aged between 20 and 54 with a chronic troublesome cough were studied in the hour after rising using audio recordings.ResultsUsing the graphical user interface (GUI), counting the number of coughs identified by HACC in an hour long recording, took an average of 1 minute 35 seconds, a 97.5% reduction in counting time. HACC achieved a sensitivity of 80% and a specificity of 96%. Reproducibility of repeated HACC analysis is 100%.ConclusionAn automated system for the analysis of sound files containing coughs and other non-cough events has been developed, with a high robustness and good degree of accuracy towards the number of actual coughs in the audio recording.

The determination of moisture in tobacco by guided microwave spectroscopy and multivariate calibration

The feasibility of using a general calibration model based on guided microwave spectroscopic (GMS) data for the determination of moisture in various tobacco types is described. Several calibration methods were used on this data; multivariate linear regression (MLR), partial least square (PLS), polynomial PLS (n-PLS) and multi-layer feed-forward artificial neural networks (MLF) and the results are compared. Promising results (approximately 2% root mean squared validation error, for moisture levels between roughly 10 and 50%) were obtained. There was no significant difference between PLS and MLF predictions. It was found that the method is sensitive to the weight of the sample used, and that best results were obtained when weight is included as a variable, whereas the position of the sample in the field had little effect. The paper suggests several experimental adaptations to improve the data and thus increases the predictive ability of models based on this data.

Improved variable selection procedure for multivariate linear regression

Abstract This paper reports the development of an improved variable selection procedure for Multivariate Linear Regression (MLR). The procedure has been compared to the more commonly applied techniques of Principle Component Regression (PCR) and Partial Least Squares Regression (PLS) and was found to outperform both techniques in terms of prediction ability of a previously unseen sample when tested using three data sets (two UV and one FT–IR data set). The technique described will illustrate that many of the shortcomings of the MLR method can be overcome by optimizing the selection of variables specifically for prediction, rather than the ability to model the training data. The paper also demonstrates that a very small calibration set consisting of the pure components only can be used to produce a good model for prediction. The procedure is iterative, and as such there are many possible combinations of variables which can be found, this paper will demonstrate that the approach will reach an optimum quickly, and give a stable answer even if the training time is short. The procedure is however more computationally time consuming than PCR and PLS but as data collection is by far the most time consuming aspect, it is not considered to be a serious problem.

Using design of experiments to select optimum calibration model parameters

A new approach to choosing the right calibration model is introduced. The basis is the well known DoE (Design of Experiments) methodology. It is shown that by identifying variables suspected to have impact on the model quality and using these as input variables in an experimental design, the significant effects and possible interactions can be determined. The chosen design has six variables: type of regression method, scaling, Box–Cox transformation, OSC pre-treatment, differentiation, and number of components. It is also shown that the approach is well suited for using more than one model evaluation criterion which is important in order to balance the fit and prediction trade-off. The feasibility of the approach is demonstrated on two different data sets. One contains visible spectra measurements of a series of metal solution standards, and the other is Raman spectra of the naphtha feed into a distillation column in a refinery. The same experimental design is used for both the laboratory and the process data in order to demonstrate the simplicity, flexibility and robustness of the proposed approach.

Chemometrical optimization FIA of perphenazine assay

A flow injection method for the assay of perphenazine using cerium(IV) as oxidant in a sulphuric acid media was adopted. Different chemometric techniques, considered the main objective of this work, were utilised to optimise sensitivity and sample throughput as a function of five experimental variables. The optimum conditions obtained by the super modified simplex method were, 0.170 M sulphuric acid, 2.13 mM cerium(IV), 2.56 ml min(-1) flow rate, 52 cm coil length, and 136 mul sample loop size. A central composite design was successfully employed, characterising the relation between these variables and the response surface, and hence validating the optimum conditions obtained by the simplex method. Regression analysis of the data from the experimental design, demonstration that a second order polynomial model is an adequate description of the surface over the factor limits studied. Further analysis of the response surface revealed the presence of a broad maximum around the simplex method optimum. A linear calibration curve in the range 50-500 ppm together with a sampling frequency of at least 120 s h(-1) and a relative standard deviation of less than 0.8% were obtained for the determination of perphenazine in its pure analytical grade. Results of the perphenazine assay in pharmaceutical preparations indicates that the method does not suffer interference from excipients rendering the method suitable for the assay of perphenazine in drug formulations.

Monitoring of the acid catalysed esterification of ethanol by acetic acid using Raman spectroscopy

The acid catalysed esterification of ethanol by acetic acid has been monitored using Raman spectroscopy. It was found that noise and baseline shift due to fluorescence spectra and process dynamics obscured the essential chemical information in the data. Principal components analysis and regression analyses have been applied progressively to de-noise the data and to extract the Raman signal. It was found that the first principal component (PC1) contained the fluorescence, while the second principal component (PC2) contained the pure Raman spectra. It was then possible to reconstruct the original data using only the second principal component, and then use this reconstructed data to follow the progress of the reaction. To be sure that the data in PC2 represented the real chemical process, the kinetics of the reaction were calculated using selected wavelength profiles in PC2. It was shown that the reaction follows theoretical second order kinetics, which is expected of this esterification.

Quantitative Fourier transform infrared spectroscopy of binary mixtures of fatty acid esters using partial least squares regression

Abstract This work describes a quantitative spectroscopic method for the analysis of binary mixtures of fatty acid esters using multivariate data models based upon Fourier Transform Infra Red (FT-IR) spectroscopy. Multivariate calibration of binary mixtures has been performed using Partial Least Squares regression (PLS), with two approaches being applied for fitting the inner relation namely a standard linear function and a polynomial function. The use of a polynomial function with PLS (polyPLS) allows what appears to be a nonlinear component in the system to be modelled effectively. Autoscaling the spectra provided the best method of data transformation for improved accuracy of prediction. The prediction abilities of the various models is illustrated using both ribbon and hexagonal plots. The percentage error in the prediction for the two PLS methods was found to be in the ranges of 4–14% and 3–9%, for the linear and nonlinear functions respectively.

Determination of acetonitrile and ethanol in water by guided microwave spectroscopy with multivariate calibration

The feasibility of using guided microwave spectroscopy (GMS) utilizing the frequency range 0.25-3.20 GHz, was combined with multivariate calibration for the determination of acetonitrile or ethanol concentration in water. A wide range of different concentrations was used (up to 30% v/v). Partial least squares (PLS) and weighted ridge regression (WRR) was applied to generate a model for prediction, based upon the microwave spectra. A high level of collinearity was observed in both of the sample data sets and this was reduced by background subtraction. The prediction ability for the two types of regression models were found to be comparable with the percentage error of prediction (PEP) being approximately 2.5% for the acetonitrile samples and 1.1% for ethanol samples.

Simultaneous Kinetic Method for the Determination of Vitamin C, Citrate and Oxalate Employing the Kalman Filter

A kinetic method for the determination of vitamin C, citrate and oxalate in their mixture is described. The method involves the use of cerium(IV) as an oxidant and measurement of reaction rates spectrophotometrically by following the decrease in absorbance of cerium(IV) at 410 nm. The adaptive Kalman filter was used for data manipulation and analysis. It is shown that the use of the Kalman filter is superior to the classical differential kinetic methods owing to its suitability for the determination of analytes that react with a single reagent and exhibit a reaction rate constant ratio of less than 1.5. The results obtained were found to be highly precise and accurate even in the presence of some expected interferents.

Multivariate calibration modelling using electrochemical/inductively coupled plasma mass spectrometry data for trace elements in ultrahigh quality water and humic acid matrices

Robust multivariate calibration is described for the determination of trace metals in water matrices. A multivariate calibration model containing electrochemical and ICP-MS concentration data for Zn, Cu, Cd and Pb mixtures in the 0.5–50 µg l–1 range was constructed using the partial least squares (PLS) method. Fifty solutions were prepared in ultrahigh quality (UHQ) water to which humic acid was added to simulate interferents. A second data set consisting of similar elemental combinations was prepared in a UHQ water matrix only. The electrochemical data was collected using anodic stripping voltammetry with ICP-MS providing independent quantitative data. All experimental work was carried out in replicate to account for variations in the ambient experimental conditions and to aid the identification of outliers. The training data set used for the calibration model was transformed prior to modelling using two separate data pre-treatment techniques. The first technique scaled the raw data using the mean of one method whilst the second used the same technique with 10% random noise added to the raw data. These two pre-treatment techniques are compared and contrasted. The calibration model using the second pre-treatment technique gave the most accurate concentration predictions for eight unknown test solutions which consisted of four solutions from either matrix. These predictions were all within a 10% relative standard error of the actual concentrations.