A comparative study for lung, colon and breast cancer diagnosis using different convolutional neural networks

• Main Authors: Akar, Hussein, Diab, Jacqueline Abou, Sbeity, Fatima, Ali, Mohamad Abou, Kassem, Abdallah, Hamawy, Lara
• Format: Conference Proceeding
• Language: English
• Subjects: baseline model; Breast cancer; CNN; Colon; colon cancer; Deep learning; histopathological images; Lung; lung cancer; Measurement; Network architecture; Safety
• ISSN: 2377-5696
• DOI: 10.1109/ICABME59496.2023.10293120

In modern times, the prevalence of diseases, particularly cancer, is rising rapidly. Diagnostic errors pose a critical safety concern in healthcare due to the complexity of medical decision-making, disease variability, and human cognitive limitations. This paper aims to compare different convolutional neural networks (CNNs) to find the most accurate model that can classify lung, colon, and breast cancer using histopathological images. A convolutional neural network is a subset of machine learning. It is one of the various types of artificial neural networks which are used for different applications and data types. A CNN is a kind of network architecture for deep learning algorithms and is specifically used for image recognition and tasks that involve the processing of pixel data. We created a baseline model that consists of a basic CNN architecture. It serves as a starting point for comparison with more complicated models. The baseline model's performance is then compared to the performance of various pretrained models (ResNet-50, ResNet-101, VGG-16, VGG-19, Inception V3, and DenseNet-121). The dataset used consists of 25,000 histopathological images of both the lung and colon and 277,000 histopathological of breast images. Results show that the ResNet-50 model is the most efficient among all the other models, scoring a validation accuracy of 98.3%. Additionally, ResNet-101, VGG-16, VGG-19, and DenseNet-121 also demonstrated strong performance, scoring 95.8%, 96.6%, 95.1%, and 92.4%, respectively. Meanwhile, Inception-V3displayed a fair result, scoring a validation accuracy of 73.2%.