The power of AI and Intel TDX: Exploring an Epilepsy Surgery Evaluation use case

iExec’s approach to confidential computing has gained recognition, being featured in Intel’s AI Inference Software & Solutions Catalogue. This article will explore the use case proof-of-concept presented by Intel, illustrating how iExec combines AI, blockchain, and Intel TDX TEEs to create a privacy-preserving healthcare use case for epilepsy surgery evaluation.

Abstract

Epileptology specializes in diagnosing and treating drug-resistant epilepsy through detailed brain analysis using electroencephalograms.

Although effective, this process is intricate and time-intensive, requiring significant, confidential data collection and complex analysis.

iExec’s research project explores solutions for these challenges by:

Creating a confidential environment for AI models to process epilepsy patients’ medical data, ensuring patient privacy and data confidentiality.
Enabling third-party use of AI models, while safeguarding model privacy when necessary.

This project utilizes Intel® Trusted Domain Extensions (TDX). It enables secure processing of confidential data and AI models while enhancing isolation and integrity.

The Context

Epilepsy is a chronic neurological disorder that affects around 50 million people worldwide. It is characterized by an enduring predisposition to generate epileptic seizures, transient occurrence of signs and/or symptoms caused by abnormal neuronal activity in the brain.

The burden of epilepsy extends beyond seizures themselves, impacting mental well-being, quality of life, and social and professional integration. The World Health Organization notes that epilepsy increases the risk of premature death by up to three times compared to the general population, with around 30% of patients experiencing drug-resistant epilepsy, where seizures persist despite anti-seizure medication. For some, epilepsy surgery may be an option.

The Challenge

Epilepsy surgery evaluation is a comprehensive process to assess if a patient is a suitable candidate for surgical intervention. It includes detailed medical history, neuroimaging studies, and electroencephalogram monitoring.

The goal is to pinpoint the epileptogenic zone — the area of the brain where seizures originate — to evaluate the surgery’s potential benefits and risks. Intracranial recordings, mapping brain activity during seizures, are crucial for planning surgery by identifying regions to target or avoid, ultimately enhancing surgical outcomes and seizure control.

Even though this method is proven, it includes significant limitations:

Data Volume and Sensitivity: An enormous amount of sensitive data has to be collected for a proper epilepsy surgery evaluation. For each patient, this can mean recording up to 256 channels, around the clock, for 1 to 2 weeks. The medical information gathered during the data collection period is also required to be kept confidential.
Analysis Complexity: Analyzing the extensive amount of collected information is time-consuming; it requires manual review by a specialized medical team. The variability in brain regions and the typology of epileptic activity across and within patients requires a high level of expertise that is not available in every hospital.

The iExec experimental approach uses Intel® TDX to:

Create a confidential environment for AI models to process epilepsy patients’ medical data, ensuring patient privacy and data confidentiality.
Enable third-party use of AI models, while safeguarding model privacy when necessary.

The Project

iExec has been at the forefront of handling confidential and sensitive information. The company has been collaborating with Intel since 2018 and is a member of the Intel Partner Alliance. This collaboration has led to building the iExec blockchain-based solution, which uses Intel® SGX enclaves to ensure confidentiality and trusted execution of computational tasks.

The fusion of confidential computing with blockchain technologies allows for secure data processing, which is especially relevant for the healthcare sector and AI. Indeed, medical information must be kept secure and confidential to avoid any data leaks and re-identification attacks. AI models are also considered strategic assets that need to be protected, as their training requires time, money, and huge amounts of data.

For this project, iExec aims to provide confidential data processing capability with Intel® Trusted Domain Extensions (TDX). TDX is Intel’s newest technology for confidential computing. Using Intel® TDX ensures patient data is processed with robust isolation, integrity, and confidentiality. The project also relies on Intel® IPEX (Intel Extension for PyTorch) to optimize Deep Learning performance on Intel processors. Intel’s IPEX integration in Python-based applications significantly enhances the performance of AI training and inferencing in TDX.

How It Works

The iExec technology combines Blockchain with Confidential Computing to ensure the security and privacy of medical data and AI models.

The Blockchain sets and records immutable governance rules about accessing patient data or the AI models, detailing who can access them, how, and under what conditions. Every interaction involving the data or the AI models is recorded as a Blockchain transaction. This gives full transparency and traceability. This also records how a model is trained, with which data, when, and by whom.

As for the Confidential Computing technology, iExec uses Intel® TDX, a hardware-based trusted execution environment. It facilitates the deployment of trusted domains, which are hardware-isolated virtual machines designed to protect sensitive data and applications from unauthorized access.

Intel® TDX ensures data and AI model integrity, confidentiality, and authenticity. It enforces the governance set on the Blockchain, making it impossible for anyone to access, copy, or tamper with the medical data to uncover patient identities or any other sensitive information.

The AI model can also be fully protected, as iExec leverages Intel® TDX to ensure the privacy of the models during the processing phase. The trained models can be encrypted by the model owners before being used, with decryption occurring only within the secure confines of an Intel® TDX confidential VM. This protection extends to preventing inspection by any party, including the model user or the Cloud provider. This opens up the way for the use and monetization of AI models by third parties without relinquishing ownership.

Furthermore, Intel’s IPEX framework deployed in the TDX environment, along with AI training and inference application, significantly enhance the AI execution performance.

Thanks to Intel® TDX and IPEX, the project aims to preserve the privacy of both patient data and AI models, even if they are used by a non-trusted third party. The goal is to expedite epilepsy surgery evaluation to allow quicker care for a greater number of patients.

‍