Automated method for scoring breast tissue microarray spots using quadrature mirror filters and support vector machines

Tissue microarray (TMA) technique is one of the widely used methods in treatment for breast cancer patients during the past decade. This technology has shown positive results in the diagnosis, detection and treatment of breast cancer. TMA spots can be classified into four main grades in which a grade of 0 indicates the spot is negative for the disease, and a grade of 3 is strongly positive. This score classification is done by pathologist and in a large scale of image data this work becomes time consuming, subjective and prone to approximate errors. The objective of this study is to find a way to classify the TMA spot images into four score types automatically and evaluate algorithm for automatic, quantitative analysis of TMA images to help pathologist save time and analyze the images accurately. This paper explores a method of automated scoring spots using density approximation of color and features clusters in the feature space, these texton histograms were then classified using multiclass support vector machines. The features used in this paper were generated by using Orthogonal quadratic mirror filters (QMF), characterized every spot by a texton histogram of nearest cluster center. The scoring performance was assessed using TMA spots from Stanford Tissue Microarray Database. The average accuracy of four classes over 50 leave-half-out experiments was around 65% to 67% with nearly balanced data, was around 58% to 60% with significant imbalanced data. The use of QMF feature of Coiflet 4 wavelet, the accuracy could be reach 80.42% for score 0; 46.78% for score 1; 64. % for score 2 and 72% for score 3.