Ability of Machine Learning and Deep Learning Models for Multiclass Classification of Kidney Stone and Lung Cancer from Computed Tomography Images: A Comparative Study

Abstract

Feature extraction is crucial in biomedical image classification because it determines the accuracy of image representations and significantly impacts the effectiveness of classification models. Deep neural network classification architectures have gained significant interest due to their ability to automatically extract important features from input images, resulting in significant progress in diverse image classification tasks in recent years. However, with the rise of deep learning techniques, traditional machine learning approaches have been largely overshadowed. This study aims to close this gap by undertaking a rigorous comparative analysis of three important machine learning models, namely Gaussian Naïve Bayes, Support Vector Machine, and Random Forest Classifier, and three advanced deep learning models, namely VGG16, InceptionV3, and Xception. The comparison is based on their ability to do multiclass classification, using two datasets kidney stone and lung cancer. Each dataset consists of four different target classes. Both machine learning and deep learning frameworks are trained separately on the datasets, with deep learning models utilizing transfer learning techniques. The performance of each model across the varied output classes is assessed using evaluation measures such as precision, recall, and F1 scores. The results of the simulation analysis reveal that both machine learning and deep learning models perform equally well, as indicated by similar F1 scores across all output classes for both datasets. This study represents a major step towards simplifying classification efforts by promoting the use of machine learning models instead of deep learning models for classifying kidney stone and lung cancer datasets. This approach helps reduce the workload and computing requirements for training.