Color-SIFT Features for Histopathological Image Analysis

Histopathology is one of the most used practices in the bio-medical field for disease and cancer detection and grading. Following the digitalization of biological data and improvement of machine/deep learning methods, the challenges for developing Computer-Assisted Diagnosis (CAD) systems arose. In this work, we explore the use of a model based on Color-SIFT descriptors, Bag-of-Features (BoF) and Support Vector Machine (SVM) to analyse and classify tumoral histopathological tissus. We tested the system using a limited amount of data and compared its results with ResNet18 results. Our model obtained a 64.8% accuracy while ResNet18 obtained 61.8% classification accuracy. We performed the experiments using the CPU rather than the GPU. The training and validation of ResNet18 took 07 H and 29 min while the proposed model took 03 H and 14 min.