Deep ensemble model with blockchain technology for lung cancer detection with secured data sharing

Lung cancer is one of the leading causes of cancer-related mortality globally, primarily due to the high frequency of late-stage diagnoses. While existing models have shown potential for enhancing diagnostic accuracy and enabling earlier detection, they may struggle with effective feature extraction from Computed Tomography (CT) scan images, and lack robust, adaptive optimization techniques. Moreover, secure and privacy-preserving sharing of sensitive medical data among healthcare providers remains a significant challenge, as conventional centralized systems pose risks of data breaches and tampering. To address this, the paper proposes a secure and efficient lung cancer detection framework. Initially, CT scan data are collected from multiple benchmark datasets. Security and privacy during data exchange are ensured through smart contracts within the blockchain network, enabling decentralized and tamper-proof data management. The lung cancer diagnosis is conducted using a novel hybrid ensemble deep learning model Hybrid Convolutional Neural Network combined with Autoencoder and Long Short-Term Memory (HCNN-ALSTM). The use of Autoencoder and LSTM in the HCNN-ALSTM approach can effectively extract the features from CT images. Further, its parameters are optimally tuned by a heuristic optimization method, the Modified Krill Herd Algorithm (MKHA). MKHA dynamically updates the global model weights to enhance learning efficiency and accuracy. Optimizing the HCNN-ALSTM framework's parameters using MKHA can effectively enhance early diagnosis accuracy from CT scan images while ensuring privacy-preserving data sharing across multiple healthcare institutions, enabling secure collaboration among healthcare providers without compromising patient privacy. Extensive experimental evaluations are performed on the proposed MKHA-HCNN-ALSTM model over existing methods. The MKHA-HCNN-ALSTM model significantly outperformed all these architectures with an accuracy of 91.64 %. Moreover, the designed framework achieved specificity of 92.2 %, Matthews Correlation Coefficient (MCC) of 84.17, Fowlkes-Mallows (FM) of 92.44, Bookmaker (BM) Informedness of 84.2, and Markedness (MK) of 84.13, which is higher than the conventional models such as Graph Convolutional Networks (GCN), Long Short-Term Memory (LSTM), Artificial Neural Network (ANN), and HCNN-LSTM model. Thus, these findings highlight the potential of the developed model for early lung cancer detection, ultimately contributing to improved clinical decision-making and patient care.