Deep Learning for Early Diagnosis of Lung Cancer

Early diagnosis of lung cancer is critical for improving patient prognosis. While Computer-Aided Diagnosis (CAD) systems leveraging deep learning have shown promise, the selection of an optimal model architecture remains a key challenge. This study presents a comparative analysis of three prominent Convolutional Neural Network (CNN) architectures InceptionV4, VGG-13, and ResNet-50 to determine their effectiveness in classifying lung cancer into benign, malignant, and normal categories from Computed Tomography (CT) images. Utilizing the publicly available IQ-OTH/NCCD dataset, a transfer learning approach was employed, where models pre-trained on ImageNet were fine-tuned for the specific classification task. To mitigate overfitting and enhance model generalization, a suite of data augmentation techniques was applied during training. It achieved an accuracy of 98.80%, with a precision of 98.97%, a recall of 96.30%, and an F1-score of 97.52%. Notably, the confusion matrix analysis revealed that InceptionV4 perfectly identified all malignant and normal cases in the test set, highlighting its clinical reliability. The study also evaluated the trade-off between diagnostic performance and computational efficiency, where InceptionV4 provided an optimal balance compared to the computationally intensive VGG-13 and the less accurate, albeit more efficient, ResNet-50. Our findings suggest that the architectural design of InceptionV4, with its multi-scale feature extraction, is exceptionally well-suited for the complexities of lung cancer diagnosis. This model stands out as a robust and highly accurate candidate for integration into clinical CAD systems, offering significant potential to assist radiologists and improve early detection outcomes.