Lung Cancer Detection System by Image Processing with Deep Neural Networks

Abstract

Lung cancer continues to be a significant worldwide health issue, mainly because it is often diagnosed at a late stage and the screening method is complicated. Conventional detection methods frequently involve intrusive procedures and heavily depend on the subjective interpretation of medical imaging by specialists. In order to improve the rates of early detection and decrease the dependence on invasive procedures, there is a crucial requirement for more precise and non-invasive diagnostic technology. This study investigates the practicality of utilizing computer-aided diagnosis tools, which are enhanced by sophisticated image processing techniques and deep learning, to enhance the identification and categorization of lung cancer from computed tomography (CT) images. The fundamental component of the proposed detection system is a deep learning model utilizing a Convolutional Neural Network (CNN), specifically engineered to examine CT scan pictures in order to identify indications of lung cancer. The model architecture consists of several layers that are specifically designed to capture different properties of the images. It starts with an initial convolutional layer that has 32 filters. This is followed by batch normalization and max pooling. This pattern is then repeated in succeeding layers, with the number of filters increasing to 64 and 128. Next, the network is flattened to establish connections with fully connected layers. Dropout is used to mitigate overfitting, and a SoftMax output layer is added for multi-class classification. The purpose of constructing the model is to gradually improve the process of learning distinctive characteristics, with the goal of capturing both basic and advanced traits that are relevant to the detection of lung cancer. The efficacy of the model was verified by employing a collection of CT scan images that were categorized into four classes according to the severity and nature of lung abnormalities. The performance was evaluated with data science criteria including accuracy, precision, recall, and F1-score. The CNN model attained a validation accuracy of 81.09%, accompanied with a validation loss of 0.64. The model exhibited exceptional precision (0.95) and recall (0.96) specifically for the most severe category of lung abnormalities, suggesting its potential usefulness in clinical environments. In summary, the findings indicate that the system based on Convolutional Neural Networks (CNN) has the ability to make a substantial impact on the timely identification of lung cancer. This system can serve as a reliable tool for radiologists and has the possibility of enhancing patient outcomes.