Robot Communication Systems; Patient Protection and Algorithmic Bias in Medical AI

The technical capstone project was a ZigBee-based robot communication system, designed for autonomous first-responding robots. In natural disasters or mass casualty incidents, the number of victims often outweighs the number of first responders; autonomous robots can help make the triaging process more efficient as they can cover a larger amount of ground in a fast time and communicate over large distances. My STS research determined how patients can be protected from bias and privacy issues when AI care is introduced in the medical system. While AI has the potential to greatly improve care and efficiency in the medical system, data privacy issues and historical health bias have the potential to be exacerbated because of it. The technical capstone inspired the research question, as leaving AI to make triaging decisions about a victim produces questions regarding fair decision-making and victim data security.
This project aids in the development of robots that have the capabilities to triage humans. The efficient and reliable communication of data in situations where there are many victims is very important. The communication system consists of a ZigBee microcontroller. This component uses radio frequencies to send data between multiple ZigBees. The other main part of the communication system includes a PCB, which was designed to power the ZigBee and handle the transfer of UART data to USB data. This system connects to a computer and a robot via a USB type A port. There are also external connections added to support sensors that transfer data using I2C or SPI communication protocols.
The results for this project were very close to the targets set at the proposal stage. The PCB successfully powered the ZigBee microcontroller, converted UART data to USB data, and transferred that data between the communication system and the “medic computer” or robot. The robots successfully communicated with each other via the design communication system. Lastly, the robots successfully received and acted on the commands to go to a predetermined goal and transfer information about the victim. The robot then waits for another command or travels back to the medic.
This research is focused on AI in the medical field and answers the following question: How can AI be developed to protect patients from bias, malpractice, and privacy concerns? To fully understand the complex implications of AI in healthcare a deep analysis was required. This research was conducted under the lens of ethics of care. The literature review was used to discover the research that had already been done to discover how patients can be protected from bias and privacy violations caused by the integration of AI into healthcare. Reviewing prior works related to this topic helped to guide the research for the discussion section.
Using the principle that obtaining objective data is a myth, this research focused on evaluating bias mitigation techniques. There are numerous methods for bias mitigation that researchers have developed, sadly while these methods have proven to eliminate the targeted bias they fail to eliminate bias as a whole from a dataset or model. Although integrating the use of AI in healthcare makes patients’ data more vulnerable, the federated learning model and edge computing are technical solutions for this issue. The federated learning model works in a decentralized manner, allowing for multiple hospitals to participate in training the same AI model while not sharing sensitive data. Edge computing can enhance the security of patient data by processing and storing that data locally, instead of transmitting it to a unified cloud location, reducing the risk of data breaches and cyber-attacks.

School of Engineering and Applied Science

Bachelor of Science in Electrical Engineering

Technical Advisor: Adam Barnes

STS Advisor: Pedro Augusto P. Francisco

Technical Team Members: Shrinidhi Nadgouda, Amelia Nist, Leila Troxell