Deep Learning-Assisted Lung Cancer Diagnosis from Histopathology Images

Abstract

Early detection plays a critical role in enhancing patient survival rates as lung cancer continues to pose a significant global health challenge and remains one of the primary contributors to cancer-related mortality. Deep learning techniques are promising as they assist doctors in disease diagnosis, especially in medical imaging. In this research, we employed a dataset comprising histopathology images of lung cancer and colon cancer from Kaggle. The data encompassed five distinct categories of the tissues of the lung and colon. To classify the images, we used a deep learning methodology that leveraged the pre-trained neural network known as AlexNet. After fine-tuning the proposed model by substituting the last fully-connected layer, we optimized the parameters using the SGDM optimizer. As a result, the overall accuracy of the method reached 99.46%. Across all considered categories, the lung benign group performed best with 100% in terms of accuracy. The overall accuracy of this research surpassed that of three previously published journal papers and six conference papers, effectively proving the remarkable capability of deep learning in accurately classifying lung cancer images. In conclusion, this research result underscores the potential of deep learning in supporting medical professionals to diagnose lung cancer.