Search

Pattern recognition in histopathological image analysis requires new techniques and methods. Various techniques have been presented and some state of the art techniques have been applied to complex textural data in histological images. In this paper, we compare the novel Adaptive Discriminant Wavelet Packet Transform (ADWPT) with a few prominent techniques in texture analysis namely Local Binary Patterns (LBP), Grey Level Co-occurrence Matrices (GLCMs) and Gabor Transforms. We show that ADWPT is a better technique for Meningioma subtype classification and produces classification accuracies of as high as 90%.

Intra-class variability in the texture of samples is an important problem in the domain of histological image classification. This issue is inherent to the field due to the high complexity of histology image data. A technique that provides good results in one trial may fail in another when the test and training data are changed and therefore, the technique needs to be adapted for intra-class texture variation. In this paper, we present a novel wavelet based multiresolution analysis approach to meningioma subtype classification in response to the challenge of data variation.We analyze the stability of Adaptive Discriminant Wavelet Packet Transform (ADWPT) and present a solution to the issue of variation in the ADWPT decomposition when texture in data changes. A feature selection approach is proposed that provides high classification accuracy.

Wavelets based analysis has been used frequently in literature for texture analysis and features extraction. Due to the availability of many wavelet filters, the issue of the selection of the optimal filter for a certain problem has always been an interesting research problem. In this paper, we present a study and comparative analysis of various wavelet filters for the problem of texture classification. The results are quite interesting as they identify wavelet properties that are desirable for wavelet based textural analysis and classification of meningioma subtypes.

The idea of multiresolution analysis has been around for over two decades now. In this paper, we explore a multiresolution analysis based technique for histopathological image classification and compare it with raw image analysis. The principle idea for the former is to derive an optimal wavelet representation, called Adaptive Discriminant Wavelet Packet Transform (ADWPT), of the images in order to achieve the largest discrimination power. Our results show that the textural features combined with ADWPT yield a significant improvement in classification accuracy.

Diagnosis and cure of colon cancer can be improved by efficiently classifying the colon tissue cells from biopsy slides into normal and malignant classes. This paper presents the classification of hyperspectral colon tissue cells using morphology of gland nuclei of cells. The application of hyperspectral imaging techniques in medical image analysis is a new domain for researchers. The main advantage of using hyperspectral imaging is the increased spectral resolution and detailed subpixel information. The proposed classification algorithm is based on the subspace projection techniques. Support vector machine, with 3rd degree polynomial kernel, is employed in final set of experiments. Dimensionality reduction and tissue segmentation is achieved by Independent Component Analysis (ICA) and k-means clustering. Morphological features, which describe the shape, orientation and other geometrical attributes, are extracted in one set of experiments. Grey level co-occurrence matrices are also computed for the second set of experiments. For classification, kernel discriminant analysis (LDA) with co-occurrence features gives comparable classification accuracy to SVM using a gaussian kernel. The algorithm is tested on a limited set of samples containing ten biopsy slides and its applicability is demonstrated with the help of measures such as classification accuracy rate and the area under the convex hull of ROC curves.