Pattern differentiation of glandular cancerous cells and normal cells with cellular automata and evolutionary learning

TL;DR: The objective of this study was to construct an autonomous feature detection system, with the hope of finding some feature patterns, based on morphological shapes (contours), that could be used to separate cancerous cells from normal cells, and showed that this system was able to search appropriate FDs to identifycancerous cells in a self-organizing manner.

Abstract: The examination of morphological features is used as a universal procedure by pathologists to determine whether cells are cancerous. Generally speaking, the shapes of normal cells are more standard (either circular or oval) than those of cancerous cells. The objective of this study was to construct an autonomous feature detection system, with the hope of finding some feature patterns, based on morphological shapes (contours), that could be used to separate cancerous cells from normal cells. A number of feature detectors (FDs) were initially generated at random. Then they were modified through evolutionary learning and cellular automata. The experimental result showed that this system was able to search appropriate FDs to identify cancerous cells in a self-organizing manner. It also showed that these FDs were general so that each of them could be used to identify more than one cancerous cell, and that there existed some common patterns of cell deformity among cancerous cells. This system was also applied to two other domains, and achieved satisfactory experimental results.