Big Data-Driven Approach for Lung Cancer Identification via Advanced Deep Transfer Learning Models

Purpose: To develop and evaluate a highly accurate computer-aided diagnosis (CAD) model based on ResNet-50 for the early identification of lung cancer-related pulmonary nodules using the publicly accessible LIDC-IDRI CT image dataset. Materials and Methods: This study utilizes the LIDC-IDRI dataset, which comprises CT scans of pulmonary nodules for the detection of lung cancer. The preprocessing pipeline involves converting all CT images to grayscale, resizing them to a consistent dimension, and applying data augmentation techniques such as rotations and flips to enhance the model's robustness. A refined ResNet-50 convolutional neural network is employed for classification to extract deep characteristics and differentiate between benign and malignant nodules. Two baseline models, the Feed Forward Back Propagation Neural Network and the Support Vector Machine (SVM), are also used for comparison to assess the efficacy of this strategy. Findings: The ResNet-50 model demonstrated superior performance across all evaluation metrics, achieving an accuracy of 99.38%, an F1-score of 99.37%, a precision of 99.91%, and a recall of 98.76%. ResNet-50 showed a high capacity to reliably detect pulmonary nodules from CT images by consistently outperforming both the SVM and the Feed Forward Back Propagation Neural Network when compared to the baseline models. Unique Contribution to Theory, Practice and Policy: Based on the findings, it is recommended that the ResNet-50–based CAD model be integrated into clinical radiology workflows to facilitate the early diagnosis of lung cancer. For broader applicability, further validation should be conducted using multi-center and prospective datasets to ensure the model’s generalizability. Additionally, incorporating real-time preprocessing and inference mechanisms within existing PACS (Picture Archiving and Communication System) platforms could streamline diagnostic processes and improve radiologist efficiency.