Enhanced Lung Nodule Classification through Ensemble Transfer Learning Models

Currently, lung cancer ranks as the second leading cause of mortality worldwide, surpassed only by heart attacks. Lung cancer is characterised by the formation of a malignant tumour in the lung, which has the potential to spread over time. The presence of this tumour can be identified in computed tomography (CT) scan images by the presence of distinct spots in the lung area. Occasionally, it might be challenging to visually identify these spots due to the inherent limitations of the human eye in perceiving subtle details. Moreover, these spots are often recognised only in the latter phases. Convolutional Neural Networks (CNNs) can effectively detect smaller abnormalities in the lung area, allowing for more detailed inspection. This paper utilises CT-scan pictures. The photos have been enhanced for classification by applying preprocessing techniques such as median filtering for noise reduction and Otsu's approach for thresholding. 3 pre-trained CNN models, including VGG16, ResNet50, and InceptionV3, have been utilised with alterations. The classification layers of these models have been eliminated and reconfigured to be trained on the particular dataset. The preprocessing processes will provide input images to the convolutional layers of these models. The output from these layers will be combined and used as the beginning layer for the newly created dense layer, which will include output nodes representing 3 classes: normal, benign, and malignant. This model achieves a remarkable accuracy of 98% on the testing dataset.