DETECTION OF LUNG CANCER THROUGH COMPUTED TOMOGRAPHIC IMAGES USING DEEP LEARNING MODELS

Abstract

Lung cancer maintains its positions as a leading cause of worldwide cancer fatalities so there remains an immediate requirement for dependable and fast diagnostic techniques. Research presents a sophisticated deep learning system which applies 5-fold cross-validation to EfficientNet-B0 for accurate CT image-based lung cancer classification. We integrated Chest CT-Scan Images and IQ-OTHNCCD datasets from Kaggle into our research which included public domain images numbering 1,415 total images. The research then utilized complex preprocessing and augmentation methods for performance enhancement. The methodology merges median noise reduction with broad data augmentation to prevent overfitting and includes EfficientNet-B0 as the classification CNN. The proposed framework reached maximum validation accuracy of 99.32% within Fold 3 when validated through 5-fold cross-method while maintaining an average accuracy of 97.43% across all folds. This method achieves superior performance compared to state-of-the-art techniques through comprehensive evaluation measurements that demonstrate precision at 97% and the other attributes at 98% and 98% and 97.9% respectively. The proposed framework demonstrates strength as an efficient method for detecting early lung cancer which shows promise to boost clinical decision-making and patient outcomes.