Abstract:
Diagnosis of anaemia is based on the observation of the variations in shape, colour and grey level distribution inside Red Blood Cells (RBC). The variations of the outer contour of RBC’s are the most important factor in the diagnosis of anaemia. As a result, only the variations in the outer contour of RBC’s are considered. Other variations, such as colours and grey level distribution, are trivial in most cases. Human being has the ability to observe features of shapes and to employ these feature in the recognition of shapes. Most of the features, from Man’s point of view, are subjective, qualitative and unmeasurable. The ability to transform these features to quantitative measurable quantities is a great success. In this paper, several qualitative features are transformed to invariant and highly noise tolerant features. Methods for measuring the new features, such as irregularity, concavity, and unsymmetry, are introduced. A new method for Run-Length coding of closed contour shapes is presented. Also, an accurate method for the computation of slope density transformation of a digitised contour is introduced. In general, most of the features discussed in this paper tend to reveal the different types of deformations occurring to any closed contour. This means that a perfect circular closed contour should yield zero values for all of the features with the exception of the area and the maximum curvature angle. Employing these new features and computation methods in haematology leads to the partial solution of the problem of the diagnosis of anaemia. In the classification step, an effective design of a decision tree is presented. In spite of the very small training set, the intellectual choice of classifiers and features used in each node lead to better generalisation, dramatic speed up of the learning process and boosting up of the classification rates.