By: Zhaoning Wang

A newly published perspective paper by Ruochen Wang, presents a thoughtful and timely overview of the growing role that generative AI—particularly ChatGPT/GPT-4—may play in the future of healthcare and biomedical research. The paper, titled “ChatGPT/GPT-4 in Healthcare: Potential Opportunities and Limitations,” was accepted in April 2024 and discusses both promising use cases and ongoing challenges in applying large language models (LLMs) to real-world medical settings.

In this concise review, Wang examines how LLMs are being tested for applications ranging from simplifying patient communication and assisting with document generation, to supporting clinical decision-making and easing the burden on healthcare professionals. The paper draws on recent studies that explore the performance of ChatGPT in answering common health questions, generating discharge summaries, interpreting radiology reports, and contributing to biomedical text mining.

“There’s been a rapid increase in interest around the use of generative AI tools in medicine, but it’s essential that we carefully consider their capabilities, as well as their limitations,” said Wang. “This paper is intended to provide a grounded overview of where these technologies are starting to prove helpful—and where further research is needed.”

Wang also highlights the model’s limitations, particularly in highly specialized tasks such as biomedical natural language processing, where the current version of ChatGPT still lags behind domain-specific models. The review emphasizes that while LLMs show potential to improve efficiency and reduce administrative overhead, their outputs must be used judiciously and with human oversight, especially when applied in clinical or research contexts.

This work adds to the growing body of literature that encourages responsible exploration of AI tools in healthcare, echoing calls for ethical use and interpretability. By summarizing early research findings, the paper helps stakeholders, from researchers to clinicians, better understand the implications of integrating ChatGPT into health systems.

As generative AI continues to evolve, Wang’s research supports informed discussion on its potential role in shaping future healthcare delivery, as well as its relevance to biomedical innovation and patient experience.

Discussion on its potential role in shaping future healthcare delivery and its relevance to biomedical innovation and patient experience.

Understanding How tDCS Impacts the Brain: New Research Offers More Realistic Modeling of Neural Responses

A peer-reviewed study co-authored by Ruochen Wang, a biomedical data scientist, provides key insights into how transcranial direct current stimulation (tDCS) affects neurons at a cellular level. The paper, published in Brain Stimulation, presents a multi-scale computational model that better simulates the effects of tDCS on specific parts of neurons, especially their axonal and dendritic structures, within realistic human brain geometry.

Transcranial direct current stimulation is a non-invasive method used to modulate brain activity and treat various neurological and psychiatric disorders, including depression and chronic pain. Despite its growing use, the mechanisms by which tDCS influences the brain at a detailed level remain unclear. Wang and his co-authors addressed this knowledge gap by combining high-resolution MRI-derived head models with biophysically realistic neuron models spanning multiple cortical layers.

“Most existing models focused only on somatic effects or used oversimplified assumptions,” Wang explains. “Our work uses morphologically realistic simulation of the neurons and shows that axon terminals may experience significantly greater polarization than the soma, sometimes up to ten times more, which has important implications for understanding both therapeutic benefits and variability in patient outcomes.”

A Key Technical Innovation: Efficient Replacement Algorithm

One of Wang’s key contributions to the project was the development and implementation of an automated replacement and rotation algorithm that ensured neuron morphologies were correctly embedded within cortical geometry, even in areas with complex anatomy like sulci and gyri. This method improved the biological realism of the simulation and reduced the error associated with unnatural placements of neuronal structures—an issue that had previously limited modeling accuracy in densely folded brain regions.

The algorithm uses fast geometric validation and a binary search–based method to optimally place and orient neurons in a way that maintains their anatomical validity without compromising simulation fidelity. This improved workflow made it possible to simulate thousands of neurons across five cortical layers efficiently, and it could be adapted for future studies using similar high-fidelity modeling approaches.

Broader Impacts

This work helps refine how researchers understand neuron-specific polarization responses during tDCS, and it contributes to more precise targeting of brain regions in both clinical and experimental settings. It may also inform the design of next-generation stimulation protocols that reduce variability in patient responses.

Beyond its clinical relevance, the modeling framework and placement method introduced by Wang can be applied to other brain stimulation modalities, such as transcranial magnetic stimulation (TMS), and supports broader efforts in computational neuroscience to simulate brain activity with anatomical precision.

Citation

Aberra, A.S., Wang, R., Grill, W.M., Peterchev, A.V. (2023). Multi-scale model of axonal and dendritic polarization by transcranial direct current stimulation in realistic head geometry. Brain Stimulation, 16(6), 1776–1791. https://doi.org/10.1016/j.brs.2023.11.018

Published by Liz SD.