Glandular structure-guided classification of microscopic colorectal images using deep learning

Abstract

In this work, we propose to automate the pre-cancerous tissue abnormality analysis by performing the classification of image patches using a novel two-stage convolutional neural network (CNN) based framework. Rather than training a model with features that may correlate among various classes, we propose to train a model using the features which vary across the different classes. Our framework processes the input image to locate the region of interest (glandular structures) and then feeds the processed image to a classification model for abnormality prediction. Our experiments show that our proposed approach improves the classification performance by up to 7% using CNNs and more than 10% while using texture descriptors. When testing with gland segmented images, our experiments reveal that the performance of our classification approach is dependent on the gland segmentation approach which is a key task in gland structure-guided classification.

Convolutional Neural Networks in the Diagnosis of Colon Adenocarcinoma

“…Although CNNs have been used for image segmentation, they originally learned only short-range spatial dependencies [50]. The segmentation approach based on transformers, which relies on self-attention mechanisms and pre-training between neighboring image patches without any convolution operations, has been demonstrated to be more efficient than CNNs [51].…”

An Investigation about Modern Deep Learning Strategies for Colon Carcinoma Grading

“…Each decoder block consists of a conv layer followed by an up-sampling layer and ReLU activation. The transformer is trained in order to detect epithelium tissue, from which colon carcinoma starts, and which hence is considered an important bio-marker for tumour detection and grading [42]. After training, it is able to provide (also on unseen visual fields) the corresponding binary masks that point out glandular regions which can be then the only part retained for subsequent processing performed by CNN models.…”

Fractal Neural Network: A new ensemble of fractal geometry and convolutional neural networks for the classification of histology images

“…Furthermore, some researchers have been recently exploring approaches that consist of ensembles of different CNN models. The authors of [5], for instance, first applied color normalization on the images of the colorectal dataset. Then, the normalized images were given as input to an U-Net CNN in order to perform segmentation, aiming to remove non-glandular areas.…”