A Wavelet approach to extract main features from indirect immunofluorescence images

Abstract

A number of previous studies have shown that IIF image analysis requires complex and sometimes heterogeneous and diversified methods. Robust solutions can be proposed but they need to orchestrate several methods from low-level analysis up to more complex neural networks or SVM for data classification. The contribution intends to highlight the versatility of Wavelet Transform (WT) and their use in various levels of analysis for the classification of IIF images in order to develop a system capable of performing: image enhancement, ROI segmentation and object classification. Therefore, WT was adopted in the de-noise section, segmentation and classification. This analysis allows frequencies characterization (low/high) and with the statistical distributions of the wavelet coefficients will be able to support the medical diagnosis process. In particular, the robustness and the goodness of the segmentation phase must be highlighted and its validation was reported in the section 3.2.1. From the depicted data it is possible to assert that the validation with ground truths produced an accuracy of 90% and that the method, to the best of our knowledge, is superior to other methods, which do not support WT (see Table 1). The advantage of using WT in all levels of abstraction of IIF data analysis lies robustness of the method, and in the rapid understanding, by the end user, of a single method that shows good average results in all levels of analysis.