Assessment of statistical and neural networks methods in NMR spectral classification and metabolite selection.

Magnetic resonance spectroscopy opens a window into the biochemistry of living tissue. However, spectra acquired from different tissue types in vivo or in vitro and from body fluids contain a large number of peaks from a range of metabolites, whose relative intensities vary substantially and in complicated ways even between successive samples from the same category. The realization of the full clinical potential of NMR spectroscopy relies, in part, on our ability to interpret and quantify the role of individual metabolites in characterizing specific tissue and tissue conditions. This paper addresses the problem of tissue classification by analysing NMR spectra using statistical and neural network methods. It assesses the performance of classification models from a range of statistical methods and compares them with the performance of artificial neural network models. The paper also assesses the consistency of the models in selecting, directly from the spectra, the subsets of metabolites most relevant for differentiating between tissue types. The analysis techniques are examined using in vitro spectra from eight classes of normal tissue and tumours obtained from rats. We show that, for the given data set, the performance of linear and non-linear methods is comparable, possibly due to the small sample size per class. We also show that using a subset of metabolites selected by linear discriminant analysis for further analysis by neural networks improves the classification accuracy, and reduces the number of metabolites necessary for correct classification.

Factors affecting the range of movement of total knee arthroplasty.

We have investigated those factors which influence the range of movement after total knee arthroplasty, including sex, age, preoperative diagnosis and preoperative flexion deformity and flexion range. We also compared cemented and cementless tibial fixation, the influence of collateral ligament and lateral parapatellar releases and of replacement of the patella, and of the period of postoperative immobilisation. We reviewed 516 Johnson-Elloy (Accord) knee arthroplasties performed between 1982 and 1989, with a minimum follow-up of 12 months. The most important factors in the range of flexion achieved after arthroplasty are the diagnosis and the preoperative range of flexion. In patients with osteoarthritis there was a mean loss of flexion; in rheumatoid arthritis there was a mean gain. In both groups, the stiffer knees gained motion and the more mobile knees lost it. Post-operative range of motion was not influenced significantly by cement fixation, collateral ligament or patellar retinacular releases, prolonged immobilisation or patellar replacement.

Some methods of combining class information in multivariate normal discrimination for the classification of human chromosomes.

We consider the use of discriminant analysis based on an assumption of multivariate Normality for allocating human chromosomes in an automated system. In this context, the assumptions which might be made about the covariance matrices for the different chromosome classes have important implications for the error rate of the system and the time required to allocate a chromosome. Linear discriminant functions based on the assumption of a common covariance matrix for all classes are fast but sometimes give bigger error rates than the assumption of a separate covariance matrix for each class. The latter assumption requires many more calculations to evaluate the associated quadratic discriminant functions. However, it is possible to assume that the covariance matrices for the different classes are, in various senses, similar to one another in order to derive other methods of combining class information on variability. These methods are here incorporated in the estimative maximum-likelihood approach to discrimination. The methods considered lead to machine classification times of human chromosomes intermediate between those for the assumptions of a common or unrelated covariance matrices. They also require the simultaneous estimation of fewer parameters than the use of a separate covariance matrix for each chromosome class. The methods are illustrated by three data sets of very different quality. Graphs of estimated error rate against classification time show that some of these ways of combining class information can be useful in the trade-off of error rate against time.