Development of Machine Learning Models for Risk Stratification and Prognostication of Advanced Lung Cancer in Computer Tomography Images

Abstract

The accurate identification of lung cancer stages is crucial for effective screening and treatment planning to extend the survival of patients. This study explores the development of machine learning models capable of accurately classifying early-stage and late-stage lung cancer patients using 800 chest computed tomography (CT) images from the dataset on Kaggle.com. The standard CRISM-DM process was used to guide data analysis and model development, aiming to extract useful features from the images that could be potential indicators of cancerous lesions. Subsequently, the performance of three machine learning algorithms, Deep Learning (DL), Artificial Neural Networks (ANN), and k-Nearest Neighbors (k-NN), was compared for the task of lung cancer stage classification. This investigation revealed that the k-NN model outperformed DL and ANN in this specific context. The k-NN model attained an accuracy of 9 6 % , Sensitivity of 9 6 . 2 5 % , and Specificity of 9 5 . 7 5 % . This research demonstrates the potential of k-NN as a beneficial and efficient tool for classifying stages of lung cancer using CT images. This advancement holds the potential to significantly accelerate diagnostic processes for medical teams and enhance patient care strategies, thereby improving the efficacy of patient management.