Secret learning for lung cancer diagnosis-a study with homomorphic encryption, texture analysis and deep learning

Area of Science:

• Medical Imaging
• Cryptography
• Artificial Intelligence

Background:

• Automated lung cancer detection from CT scans is crucial for early diagnosis.
• Deep learning models show promise but raise privacy concerns due to data sharing.
• Existing privacy methods like federated learning and differential privacy have limitations.

Purpose of the Study:

• To propose a novel method for privacy-preserving lung cancer classification using homomorphic encryption.
• To address the limitations of existing privacy-preserving techniques in medical imaging analysis.

Main Methods:

• Implementation of homomorphic encryption on CT scan images (normal, adenocarcinoma, large cell carcinoma, squamous cell carcinoma).
• Textural information extraction from encrypted data.
• Classification of encrypted data using deep learning models.

Main Results:

• Achieved a classification accuracy of 0.9347 for lung cancer detection.
• Demonstrated the feasibility of performing computations on encrypted medical data.

Conclusions:

• Homomorphic encryption offers a robust solution for privacy-preserving deep learning in medical image analysis.
• The proposed method effectively classifies lung cancer types while maintaining data confidentiality.