Detection of lung cancer types using CT scans

Abstract

Lung cancer is a devastating global health issue that can be fatal if not detected early. To increase the chances of successful treatment and prevent the loss of lives, doctors need to identify the type, which the lung cancer belongs to. Currently, CT scans are commonly used in medical practice to detect and diagnose lung tumors. However, implementing deep learning models to identify the lung cancer types poses a significant challenge. Because acquiring many medical images for each type of lung cancer can be difficult. The effectiveness of deep learning algorithms which were implemented for image recognition relies on the diversity of data samples. However, creating a comprehensive dataset of significant size requires a significant amount of human effort for manual labeling, which makes it a time-consuming process. The challenge of requiring a large amount of data samples for each category in traditional deep learning models was addressed in this research by implementing a prototypical network, which is a few- shot learning technique. This method requires only a few samples per category, and it was used in conjunction with a pre-trained model to extract features from lung CT scans. The accuracy of the model was analyzed based on the number of samples per category. Overall, the results of the study demonstrate that implementing a prototypical network for lung cancer type detection is feasible. Human interpretation of medical images can vary among different medical professionals. Therefore, this approach can act as a decision support tool for medical professionals which makes it more accessible and cost-effective.