Cross Organ Bridge Transfer Learning for Lung Cancer Detection

Classification of medical CT scan images is a challenging task that requires high accuracy and sensitivity. In this paper, we propose a method based on VGG 19 and transfer learning for the classification of normal and tumor of lungs from CT scan images. A popular deep convolutional neural network for image identification applications is VGG19. VGG 19 is used as a feature extractor and the last layers are fine-tuned to adapt to our task. We also use transfer learning to leverage the knowledge learned from other datasets, such as chest CT images for different types of lung cancer, viz. adenocarcinoma, basal cell carcinoma, squamous cell carcinoma and kidney dataset. We use a public dataset to assess our methodology. Our findings demonstrate the great accuracy and sensitivity of our approach. Using CT scan pictures, our technique may be helpful for the early identification and diagnosis of lung cancer. In the area of medical imaging, convolutional neural networks (CNNs) in particular have shown promising results in picture categorization tasks. Deep learning techniques, however, face difficulties because of the medical domain's lack of appropriate labelled data. In our study, we provide a novel transfer learning-based approach to medical picture categorization. By using a bridged database created using the same medical imaging modality as the target database, we utilize modality-bridge transfer learning to overcome the domain imbalance between the source and target datasets. Even with a small number of labelled target medical pictures, we are able to obtain good classification accuracy by learning from the source to the bridge and then from the bridge to the target. In order to extract deep characteristics and provide highly accurate findings, our research uses VGG19, a popular deep convolutional neural network, to improve the classification performance of pictures of lung cancer.