Lung Cancer Diagnosis and Classification Using Hybrid Deep Feature Extraction

Lung cancer, a leading cause of mortality, necessitates prompt detection for improved patient outcomes. Recent strides in diagnostic technologies, including biomarkers, genetic testing, and advanced imaging like CT and MRI, have revolutionized lung cancer detection. Computer-aided diagnosis (CAD) tools offer invaluable insights, rapidly analyzing imaging data for early identification. Integration of machine learning, deep learning, and CNNs has further transformed lung cancer diagnosis, empowering pulmonologists with highly accurate and efficient tools for interpreting medical imaging. Feature extraction plays the most vital role in the detection of lung cancer. In this research, a hybrid approach is proposed for better feature extraction, and then different classifiers are used to classify it for efficient detection of lung cancer. The dataset is taken from IQ-OTH/NCCD and is publicly available on Kaggle. Median and Gaussian filters are used for data preprocessing. To enlarge the dataset and avoid the chances of overfitting, data augmentation is used. Then, feature extraction is done on the preprocessed dataset using handcrafted methods HOG, GLCM, and pretrained automated models DenseNet201 and VGG-16. Features extracted from these algorithms are concatenated into a single array. Different combinations are tested with SVM, XGBoost, Artificial Neural Network (ANN) classifiers, and Logistic Regression. Results are compared in accuracy, precision, recall, and score. The results obtained from these combinations after classifying them with SVM, ANN, XGBoost, and Logistic regression give efficient results, and one of the combinations outperforms the previous results, providing an accuracy of up to 99.3% after being classified with SVM.