Abstract
The field of radiology is one that is rapidly evolving, with the integration of artificial intelligence (AI) and machine learning, medical devices now have the power to shape diagnosis, intervention, workflow, and patient outcomes. My two projects this past academic year closely examine this landscape from complementary perspectives: one being an STS analysis of the ever-growing complex relationship of AI-enabled radiology devices and its regulation, whilst the other technical project focused on improving needle safety, specifically for interventional radiology. These two projects together address a larger global concern, as radiology devices become more sophisticated and dependent on AI and machine learning algorithms, how will the systems that govern its design, approval, testing, and relevance keep pace with the rising clinical and ethical risks that they introduce.
STS Research Paper
My STS project dissects and evaluates whether the U.S. Food and Drug Administration’s (FDA) current regulatory framework and guidelines are sufficient enough to oversee AI-enabled radiology devices. By using Actor-Network Theory (ANT) as the primary framework and meta-analysis, the paper argues that these technologies are not just tools, but a part of a complex sociotechnical network that involves regulators, manufacturers, clinicians, patients, datasets, algorithms, approval pathways, and clinical workflows. With this framework in mind, the paper first compares FDA regulation with the European Union’s Medical Device Regulation (EU MDR) to understand the difference in oversight. My analysis shows that the FDA’s current oversight framework, specifically with 510(k) clearance and Predetermined Change Control Plans, are suited only to previously static medical devices, rather than the adaptive evolving AI systems that are being approved today. In addition, this paper explores and highlights major concerns about AI-enabled medical devices, like post-market post-market surveillance, recall risk, algorithmic bias, limited demographic reporting, and weak transparency in validation studies. The paper, in turn, argues the need for stronger lifecycle monitoring, mandatory performance reporting, and an overall more robust oversight of post-approval changes. Within this argument and perspective, AI-enabled radiology devices are not simply just technical innovations, but are independent actors that can reshape healthcare, influence clinical decision making, reinforce disparities, and create new legal and ethical complications if regulatory systems remain outdated.
Technical Report
My technical project addresses a more specific radiology problem, the risk of needlestick injuries (NSIs) during interventional radiology (IR) procedures. IR procedures often require repeated needle withdrawal and reinsertion, as clinicians perform aspiration, tissue sampling, guidewire placements, or other image-guided interventions. The project team, myself included, developed and iterated prototype designs for a sheath-based safety system aiming to reduce NSIs whilst still preserving clinical usability and relevance. The project moved through multiple CAD iterations and first-generation 3D-printed designs, including a telescoping sheath design and a mechanical-friction attachment design. The early stages of testing focused on feasibility, the locking mechanism and structural integrity. The results of testing were promising, showing that the prototype could function as a relevant safety system, and with material refinements could improve the performance of the device. Although the team encountered design challenges related to hub compatibility and printing access, the project was able to establish a viable design and foundation for further iterations and testing with ultrasound visibility and durability assessments.
