Usare l’AI per superare le disuguaglianze di salute nel bambino: Alberto Tozzi

Healthcare resources are not distributed equitably worldwide, affecting pediatrics and other fields.

Access to health varies by geography and sometimes by postal code within the same city.

Focal cortical dysplasia is a rare disease that causes epilepsy; children with this condition remain with seizures insensitive to treatment until someone recognizes the condition, which can be resolved completely with surgery.

Algorithms can be trained to improve access to diagnosis and care.

It is faster to transmit a diagnostic algorithm to a remote location than to send a specialist.

Training fair algorithms requires data representing all populations that will use the technology.

More training data improves algorithm performance.

Accessing patient health data in hospitals and public structures is very difficult, especially in Europe.

GDPR regulation, while protecting patient privacy, is slowing potential direct benefits to patients.

Smartphones use local AI models for voice recognition without a central global database of all users' voices.

In federated learning, local model parameters are shared centrally with significant data savings and without transmitting personal data, thus preserving privacy.

Federated learning architecture is highly feasible, especially using simple intermediaries like smartphones attached to MRI machines as network nodes.

One difficulty in advancing randomized clinical trials is identifying and recruiting eligible candidates.

AI agents can be taught eligibility criteria for a protocol and placed in a hospital's electronic health record to screen 24/7 for potential study candidates.

Federating AI screening agents across institutions can make clinical trials significantly more efficient by accelerating patient identification and recruitment.

Patients often search online for help identifying the best therapy for a tumor, rare disease in their child, or personal problem, and this can be addressed with the same AI agent solution.

Anthony Chang is a cardiologist in California who received a consultation request from a Southeast Asian country where an echocardiographer could not diagnose a patient and could not transmit the echocardiography video due to technical issues.

The child in the Southeast Asian case died; at the operating table the child had characteristics of a complex cardiac condition that could have been diagnosed with an AI algorithm.

After this case, Anthony Chang enrolled in a Stanford AI course and worked with Stanford engineer Tim Chu to build models supporting equitable care.

The Pediatric Moonshot project aims to support reduction of care inequalities because all children and adults have a right to health, and technology can help achieve this.