An algorithm for cancer nest feature extraction from pathological images

Here, we propose an algorithm to automatically obtain extraction filters for the affected regions from cancer images. The proposed algorithm consists of two steps: extraction of affected region candidates and elimination of false positives. Useful features of cancer images, such as the area and degree of circularity of cancer nests, etc., are extracted using the derived filters. These features are useful for supporting pathological diagnosis. Automatic Construction of Tree-structural Image Transformation (ACTIT) was used to construct these filters to extract the affected regions from the image. The proposed algorithm was applied to a mouth cancer pathological image. The results confirmed that the proposed algorithm can obtain good filters capable of extracting cancer nests. The derived filters were also applied to other images from the same specimen. The results also indicated that the generated filters show general versatility in extracting cancer nest candidates. The area and degree of circularity of the cancer nets were also derived automatically.