Abstract:
Analysis of Histopathological images for cancer detection and grading has an important role in clinical medicine. Manual analysis of cellular images depends on experience of pathologist and is a very time-consuming and hectic task for human being. Machine learning based computer aided solutions for this problem have helped pathologists which comes in category of biomedical image processing. Algorithms are available for analysis and detection of clustered nuclei in histological images, their shapes, grading, count and for their classification as benign or malignant. Clustering algorithms are applied to rarely available dataset of histological stained images thus their efficiency cannot be evaluated properly. To address this issue, a comprehensive dataset is required having different types of cellular images to assess the performance of any clustering algorithm. Aim of this study is to generate a synthetic dataset of different types of colored cellular images with their ground truth. Clustering algorithms can be applied on any image from dataset and its performance can be evaluated. For synthetic images generation, a small dataset of real images of Renal Cell Carcinoma is used provided by supervisor. Dataset include 4 different type of cellular images. To generate the image, features of real images are extracted and randomized to get a new image on every new simulation run. We have used 16 images for data set i.e. 4 of each type and generated 4 different images from every input image. As a result, we have generated a data set of 64 synthetic images with their ground truth.
