GroupMixer: Patch-based Group Convolutional Neural Network for Breast Cancer Detection from Histopathological Images

Ardavan Modarres,Erfan Ebrahim Esfahani,Mahsa Bahrami

2023-11-16

Abstract:Diagnosis of breast cancer malignancy at the early stages is a crucial step for controlling its side effects. Histopathological analysis provides a unique opportunity for malignant breast cancer detection. However, such a task would be tedious and time-consuming for the histopathologists. Deep Neural Networks enable us to learn informative features directly from raw histopathological images without manual feature extraction. Although Convolutional Neural Networks (CNNs) have been the dominant architectures in the computer vision realm, Transformer-based architectures have shown promising results in different computer vision tasks. Although harnessing the capability of Transformer-based architectures for medical image analysis seems interesting, these architectures are large, have a significant number of trainable parameters, and require large datasets to be trained on, which are usually rare in the medical domain. It has been claimed and empirically proved that at least part of the superior performance of Transformer-based architectures in Computer Vision domain originates from patch embedding operation. In this paper, we borrowed the previously introduced idea of integrating a fully Convolutional Neural Network architecture with Patch Embedding operation and presented an efficient CNN architecture for breast cancer malignancy detection from histopathological images. Despite the number of parameters that is significantly smaller than other methods, the accuracy performance metrics achieved 97.65%, 98.92%, 99.21%, and 98.01% for 40x, 100x, 200x, and 400x magnifications respectively. We took a step forward and modified the architecture using Group Convolution and Channel Shuffling ideas and reduced the number of trainable parameters even more with a negligible decline in performance and achieved 95.42%, 98.16%, 96.05%, and 97.92% accuracy for the mentioned magnifications respectively.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The paper aims to address the issue of diagnosing breast cancer at early stages, particularly in detecting malignant breast cancer from histopathological images. The authors have introduced a new architecture called "GroupMixer," which is a Patch-based Group Convolutional Neural Network, for detecting the malignancy of breast cancer from histopathological images. Traditionally, histopathologists manually analyze these images to look for malignant patterns, but this task is both time-consuming and requires expertise, and may not be immediately available in many cases, leading to extended waiting times for patients and potential impacts on them.

To accelerate the diagnostic process and reduce human errors, researchers have turned to machine learning and computer-aided diagnostic systems. In recent years, deep learning methods, particularly Convolutional Neural Networks (CNNs), have made significant progress in automatically extracting the information features needed for classification, without relying on domain experts' manual feature extraction knowledge. However, despite CNNs' dominance in the field of computer vision, Transformer-based architectures have also shown promising results in different computer vision tasks. Nevertheless, Transformer-based architectures are typically larger, with a significant number of parameters, and require large datasets for training, which is not common in the medical field, especially for breast cancer datasets.

The method mentioned in the paper combines the idea of a fully convolutional neural network architecture with patch embedding operations, proposing an efficient CNN architecture for detecting the malignancy of breast cancer from histopathological images. This approach, despite having significantly fewer parameters than other methods, achieved accuracies of 97.65%, 98.92%, 99.21%, and 98.01% under 40x, 100x, 200x, and 400x magnifications, respectively. Furthermore, by utilizing the concepts of Group Convolution and Channel Shuffling, the number of trainable parameters was further reduced, while the performance drop was negligible, achieving accuracies of 95.42%, 98.16%, 96.05%, and 97.92% at the same magnifications, respectively.

Overall, the "GroupMixer" architecture proposed in the paper not only effectively identifies the malignancy of breast cancer but also has far fewer parameters than other methods, holding the potential for deployment on low-cost embedded devices. This is a significant contribution for fields lacking large-scale medical datasets.