Optimizing Knowledge Transfer in Sequential Models: Leveraging Residual Connections in Flow Transfer Learning for Lung Cancer Classification
Richa Sharma, Santosh Kumar, Abhishek Shrivastava, Tejasv Bhatt
Abstract
Lung cancer is a prevalent and life-threatening disease characterized by abnormal cell growth in lung tissue. Early detection and accurate classification of lung cancer are crucial for timely treatment and improved patient outcomes. Our work proposes, a novel framework for accurate multi-class categorization of lung cancer using deep learning. The proposed framework uses a customized Densenet-201 model to leverage the knowledge of transfer learning, which acts as a parent architecture, enhanced by a residual structure as the child architecture. The performance of our model was examined by conducting experiments on the LCS25000 data set. The results demonstrate the outstanding accuracy of our proposed framework, achieving a remarkable accuracy of 95% on the test data set. This signifies the model’s ability to accurately classify different types of lung cancer based on histopathology images. Our model also achieved remarkably well results in the TCGA lung cancer dataset, hence proving its generalization. These findings have important repercussions for improving pulmonary pathology diagnostic abilities and hold promise for enhancing patient care in the field of lung cancer diagnosis.