From ChatGPT to DeepSeek: Potential uses of artificial intelligence in early symptom recognition for stroke care

Abstract

In the era of artificial intelligence (AI), healthcare and medical sciences are inseparable from different AI technologies[1]. ChatGPT once shocked the medical field, but the latest AI model DeepSeek has recently taken the lead[2]. PubMed indexed publications on DeepSeek are evolving[3], but limited to editorials and news articles. In this Letter, we explore the use of DeepSeek in early symptoms recognition for stroke care. To the best of our knowledge, this is the first DeepSeek-related writing on stroke. In 2023, our group published here a patient’s experience of using ChatGPT to guide his early medical attendance[4]. With the availability of DeepSeek, the same set of questions were fed for response evaluation. DeepSeek’s outputs are presented in Table 1. A step-by-step guide was yielded, the acronym of “FAST” in stroke management was mentioned. Towards the end, the responses of the chatbot were very humanized, even with a “blue heart” emoji, providing psychological care and comfort to the users. Similar to ChatGPT, DeepSeek is also a reliable tool for early symptom recognition.Table 1: Chat record with DeepSeek on 15th February, 2025. The acronym of “FAST” in stroke management was also mentioned by DeepSeek-R1.DeepSeek, as an AI assistant, plays a supportive and educational role in pre-hospital stroke care. While it is not a substitute for professional medical care, it can contribute in symptom recognition, checklist tools, triage support, and more, to shorten time to treatment. For caregivers, DeepSeek also helps create clinical scenarios to maids and first responders to rehearse on what actions to take upon recognizing stroke patients. The checklist generated by DeepSeek includes 7 different categories: facial drooping, arm weakness, speech difficulty, sudden vision changes, balance or coordination issues, sudden severe headache, confusion or cognitive changes. Within each category there is a stem [e.g. Are you experiencing a sudden, severe headache with no known cause (like a ‘thunderclap’ headache)?] with 2 options: “Normal” versus “Concern”. These question-and-answer-based checklists are quite user-friendly, and practical in the environment of emergency. Simulation is an important pillar in medical education, particularly in team-based practice like emergency medicine[5]. To facilitate learning, debriefing via different methods after the simulations is essential for participants to gain and grow[5]. Concerning scenario-based simulation, we believe DeepSeek’s performance was non-inferior to our traditional medical textbooks. DeepSeek also provides a post-simulation debriefing, with different levels of difficulties and learning objectives. Here is one of the examples: You’re at home with your father. He suddenly slumps in his chair, drops his coffee cup, and says his words are ‘coming out wrong.’ What do you do next? … It’s been 5 minutes since symptoms started. What do you do? … Post-simulation debrief Key takeaways: FAST criteria (Face, Arms, Speech, Time). Importance of immediate action (time=brain cells saved). Common mistakes: Waiting to see if symptoms improve. Transporting the patient without professional help. Real-world stats: Patients who arrive at the hospital within hours of symptoms have better recovery rates. In short, the role of chatbots in medical emergency pre-hospital care is evolving. DeepSeek is quite useful for early symptom recognition, as well as caregivers training. Conflict of interest statement The authors report no conflict of interest. Funding This study received no extramural funding. Authors’ contributions Concept or design: SCLA; Acquisition of data: WYL; Analysis or interpretation of data: WYL, SCLA; Drafting of the article: WYL; Revising the manuscript: SCLA. Publisher’s note The Publisher of the Journal remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Edited by Liu JJ

Ähnliche Arbeiten