Cancer Identification in Enteric Nervous System Preclinical Images Using Handcrafted and Automatic Learned Features

Chronic degenerative diseases affect Enteric Neuron Cells (ENC) and Enteric Glial Cells (EGC) in shape and quantity. Thus, searching for automatic methods to evaluate when these cells are affected is quite opportune. In addition, preclinical imaging analysis is outstanding because it is non-invasive and avoids exposing patients to the risk of death or permanent disability. We aim to identify a specific cancer experimental model (Walker-256 tumor) in the Enteric Nervous System (ENS) cells. The ENS image database used in our experimental evaluation comprises 1248 images taken from thirteen rats distributed in two classes: control/healthy or sick. The images were created with three distinct contrast settings targeting different ENS cells: ENC, EGC, or both. We extracted handcrafted and non-handcrafted features to provide a comprehensive classification approach using SVM as the core classifier. We also applied Late Fusion techniques to evaluate the complementarity between feature sets obtained in different scenarios. In the best case, we achieved an F1-score of 0.9903 by combining classifiers built from different image types (ENC and EGC), using Local Phase Quantization (LPQ) features.