Exploring Deep Learning and Machine Learning Techniques for Histopathological Image Classification in Lung Cancer Diagnosis
Rashadul Islam Sumon, Md Ariful Islam Mazumdar, Shah Muhammad Imtiyaj Uddin, Hee‐Cheol Kim
Abstract
To diagnose lung cancer, this study thoroughly investigates the deep learning and machine learning methods used to classify histopathological lung cancer images. Using Dense-Net 121 for feature extraction, the study examines the effectiveness of several machine learning algorithms in this crucial medical task, including Random Forest, Support Vector Machine (SVM), AdaBoost, artificial neural networks, K-Nearest Neighbors (KNN), and XGBoost. Most remarkably, the SVM algorithm is the most effective, with remarkable precision in training and testing. With a testing accuracy of 96.6 percent and a training accuracy of 99.99%, the SVM model specifically trained on features retrieved by Dense-Net 121 exhibits outstanding performance. Additionally, the SVM classifier demonstrates remarkable recall (96.40%), F1 score (96.39%), Cohen Kappa score (94.61%), and precision (96.39%), demonstrating its efficacy in correctly classifying histological images suggestive of lung cancer. These results highlight the possibility of combining SVM with deep learning-based feature extraction to improve lung cancer detection efficiency and accuracy, improving patient care and clinical decision-making.