Precise Lung Cancer Prediction using ResNet – 50 Deep Neural Network Architecture

Abstract

The fact that lung cancer continues to be the leading cause of cancer-related death around the world emphasizes how important it is to improve diagnostic methods. Using computed tomography (CT) images and deep learning techniques, the goal of this study is to improve the classification of lung cancer. EfficientNetB1 and Inception V3 are two well-known convolutional neural network (CNN) architectures that we compare the performance of our modified ResNet50 architecture against in order to determine how well it performs in the classification of lung nodules. Analyzing the effects of various preprocessing and hyperparameter optimization methods on model performance is one of our research objectives. Another is to determine how well these models improve diagnostic accuracy. An extensive collection of CT images with annotated lung nodule classifications make up the utilized dataset. To ensure accurate model training and improve image quality, a rigorous preprocessing pipeline is used. Using the Keras Sequential framework, the models are trained with optimal dropout rates and L2 regularization to prevent overfitting. Metrics like accuracy, loss, and confusion matrices are used to evaluate model performance. A comprehensive evaluation of the model's sensitivity and specificity across various thresholds is also provided by means of the Free-Response Receiver Operating Characteristic (FROC) curve and Area Under the Curve (AUC) values. The adjusted ResNet50 model showed prevalent order exactness, accomplishing a precision of 98.1% and an AUC of 0.97, in this way beating different models in the review. EfficientNetB1 had an accuracy of 96.4 percent and an AUC of 0.94, while Inception V3 had an accuracy of 95.8 percent and an AUC of 0.93, as a comparison. Based on these findings, it appears that the accuracy of lung cancer detection from CT images can be significantly improved by combining specialized preprocessing and training methods with advanced CNN architectures. With potential implications for clinical practice and future research directions, this study offers a promising strategy for increasing lung cancer diagnostic accuracy.