A feature selection based framework for histology image classification using global and local heterogeneity quantification.

Biopsy remains the gold standard for the diagnosis of chronic liver diseases. However, the concordance between readers is subject to variability causing an increasing need of objective tissue description methods. A complete framework has been implemented to analyze histological images from any kind of tissue. Based on the feature selection approach, it computes the most relevant subset of descriptors in terms of classification from a wide initial list of local and global descriptors. In comparison with equivalent methods, this implementation is able to find lists of descriptors which are significantly shorter for an equivalent accuracy and furthermore it enables the classification of slides using combinations of global and local measurements. The results have pointed that it could reach an accuracy of 82.8% in a human liver fibrosis grading approach by selecting 6 descriptors from an initial set of 258 global and local descriptors.

Contrast enhanced ultrasound image segmentation based on fuzzy competitive clustering and anisotropic diffusion.

Doppler has been used for many years for cardiovascular exploration in order to visualize the walls and anatomical or functional diseases. The use of ultrasound contrast agents makes it possible to improve ultrasonic information. Thus, harmonic imaging is a powerful tool for characterizing pathologies by their vascularity. However, images are buried in noise. We present an approach for noise reduction and structural features enhancement based on nonlinear anisotropic diffusion, coupled with fuzzy clustering segmentation to delineate liver pathologies.