(US & Canada) | Balancing Data Privacy and Model Training is Like a Delicate Dance — GE HealthCare Senior VP and General Manager

Vignesh Shetty, Senior VP and General Manager, GE HealthCare, speaks with Jack Berkowitz, Chief Data Officer at Securiti, in a video interview, about his role at GE, how the organizational AI platform aids data modernization, centralizing data versus making data actionable where it is, achievements with AI, challenges around data protection, and facilitating AI adoption.

GE HealthCare is a stand-alone company that offers medical technologies, including imaging, ultrasound, healthcare IT, contrast media, molecular imaging agents, and more.

The organization is driven to create a world where healthcare has no limits, says Shetty. He considers his role as a bridge builder across teams and customers working across the data and AI infrastructure.

Expanding, Shetty mentions building foundational capabilities such as data platforms, AI tooling, and identity management, which lay the groundwork to harness the data and AI potential.

In addition, he partners with customers, both internal and external to GE HealthCare, through the technology and tools his team builds. This accelerates the building, deployment, and monitoring of smart devices.

Commenting on where the organization stands in terms of data modernization, Shetty stresses that 30% of the world's data is generated by the healthcare industry. When it comes to the current state of data estate, he notes that, for the first time, it is not just raw information but a treasure trove of insights waiting to be unlocked.

Shetty further states that the organizational AI platform is a modern factory that takes those raw materials and converts them into actionable insights or products. These, in turn, create value for both patients and healthcare providers at large.

Moving forward, Shetty shares that the organization is challenging the traditional approach to data management, which involves breaking down silos and aggregating data into a central repository. Instead, he believes that it will be more effective to design systems that make data actionable where it is rather than bring it to one central location.

Speaking of achievements, Shetty maintains that from the customer standpoint, it would be the ability to simplify workflows and drive effectiveness in addition to efficiencies. This could be done by leveraging technology.

For instance, he mentions how the teams have partnered to build the AIR Recon DL, which is a deep learning algorithm embedded into the GE MRI machines. Shetty affirms that, compared to the past, the image quality has improved and the time taken for an MRI has reduced by up to 50%.

This has created a virtuous flywheel, making hospitals happy with increased productivity and better patient experiences, he adds. Shetty hopes that it will also enhance customer affordability.

Parallelly, he takes pride in how the company has democratized the ability to build, deploy, and monitor AI across organizations. It is similar to bestowing superpowers on every employee, where they can multiply collective intelligence while adhering to AI principles.

Thereafter, Shetty comments on the challenges around data protection. He states that both the security and privacy of patient data and other data must be earned. He says that while training the employees, they are educated that data protection should be a shared sense of responsibility while building a product.

Balancing data privacy with model training is critical, and the organization uses techniques like de-identification, data anonymization, and differential privacy to ensure training AI models without compromising patient confidentiality.

Furthermore, Shetty believes that this educational practice should extend to customers and patients to achieve the security goal and be responsible even while investing in the most secure technology.

To facilitate seamless AI adoption, he states that the AI systems must be trustworthy. One major obstacle to this is ensuring the accuracy and fairness of the models used.

In conclusion, Shetty states that to mitigate the obstacle, organizations must monitor model performance, examine bias, use synthetic data, encourage data sharing for validation, and emphasize explainable, casual, and ethical AI.

CDO Magazine appreciates Vignesh Shetty for sharing his insights with our global community.