Efficient Bag of Deep Visual Words Based features to classify CRC Images for Colorectal Tumor Diagnosis
Kamal Saluja, Ankit Bansal, Amit Vajpaye, Sunil Gupta, Abhineet Anand
Abstract
CRC images are one of the most essential diagnostic techniques for colorectal cancer detection. The bag of deep visual words-based methodology (BoDVW) has proven to be a significant representation of CRC images for its enhanced discriminability when compared to earlier deep learning-based methods. This study presents a new Efficient Bag of Deep Visual Words (EBoDVW) feature that takes advantage of three alternative scales of the VGG16 model's 4th pooling layer's output feature map. Max pooling operation has done with three kernels for EBoDVW-based features. The proposed characteristics are evaluated using the Support Vector Machine (SVM) classification technique on the CRC public dataset, which contains over 1133 CRC pictures. The results of proposed experiments reveal that our strategy generates a consistent and prominent classification accuracy of 98.2 %.