Segmentation and classification of biological cell images by a multifractal approach

Abstract

Recently, the clinical role of image processing has been developed considerably. The resources of this new technology were exploited for the needs of doctors in their practice. In this study, we propose a computer vision for tracking the uterine collar cancer. Here, we present three stages: preprocessing, segmentation, and classification. The segmentation stage uses a multifractal algorithm based on the computation of the singularity exponents; its role is separating each cell on its core and its cytoplasm, which permits the analysis of each one in the recognition stage for deducing a response about the malignity of the cell. However, the classification is performed by an algorithm of area growth. Knowing that there are four layers in the epithelium, the classification allows for learning the type of each cell in an image for organizing the research in the recognition stage. Thus, we contribute to the creation of a database for the recognition stage. This base contains the core and cytoplasm images with information about the type of each cell. Promising results were obtained with a short execution time that permits the start of the recognition stage.

DeepEC: An error correction framework for dose prediction and organ segmentation using deep neural networks

“…Medical image segmentation tasks are very similar to the dose prediction task, and DCNNs have been applied to many tasks of this field. 6,8,9,17 Compared with traditional methods, [18][19][20] DCNN-based methods can extract different levels of features automatically, from low to high, as the networks become deeper in a data-driven manner, which makes it no longer necessary to manually define features, thereby reducing the workload for doctors. Because manually delineating OARs based on CT images for radiation therapy is time-consuming and error-prone.…”