Enhancing Privacy-Preserving Cancer Classification with Convolutional Neural Networks.

Area of Science:

• Computational biology
• Genomics
• Machine learning

Background:

• Precision medicine improves cancer prognosis through personalized treatments.
• Accurate cancer classification, especially for metastatic cases, relies on primary tumor location.
• Current methods are time-consuming and costly, necessitating automated solutions.

Purpose of the Study:

• To introduce OGHE, a novel deep learning approach for privacy-preserving cancer classification.
• To leverage Homomorphic Encryption (HE) for secure processing of sensitive genomic data.
• To enhance classification accuracy and efficiency in cancer diagnostics.

Main Methods:

• Utilized a convolutional neural network (CNN) architecture for genomic data analysis.
• Implemented Homomorphic Encryption (HE) to ensure data confidentiality during computation.
• Developed VarScout, a feature selection method preserving spatial patterns for improved accuracy.
• Incorporated an efficient packing mechanism to mitigate HE computational overhead.

Main Results:

• OGHE demonstrated effective privacy-preserving cancer classification on the iDash 2020 dataset.
• The packing mechanism significantly reduced latency compared to existing solutions.
• VarScout successfully identified significant features while preserving spatial genomic patterns.
• Achieved accurate cancer classification with guaranteed data confidentiality.

Conclusions:

• OGHE offers an effective and efficient solution for privacy-preserving cancer classification.
• The integration of HE and CNNs provides a robust framework for secure genomic data analysis.
• This approach has the potential to advance precision medicine by enabling secure utilization of sensitive patient data.