思想领袖
当AI进入医院时：医生和病人有什么变化
在 Google 上的首选来源中添加 Unite.AI
在前一栏，我从商业角度探讨了医疗AI，重点是它为技术公司创造的机会。 但除了创新和新市场的潜力之外，还有一个更根本的问题：这些技术对使用这些技术的医生和依赖这些技术的病人有什么区别？毕竟，医疗创新的真正价值在于其提高护理质量和获得护理的能力。 我已看到如何提供护理的显著变化。
工作人员短缺如何加速保健自动化
我认识一个企业家，他的公司为医院提供临时医务人员，包括护士，他们在没有正式雇员时负责工作。 你可能会期望这样的企业在人员持续短缺的情况下蓬勃发展。 但是，他称，甚至那些专门提供保健人员的公司也在努力寻找足够的合格人员。
几十年来，医院一直面临医生和护士短缺的问题。 培训新的保健专业人员需要大量的时间和资源。 医学教育，特别是美国的医学教育，可能过于昂贵，学习和临床训练的年代伴随着经济和心理压力。 即使在完成培训之后，医生仍然面临高度的压力、巨大的责任和持续的时间限制。
因此，医生有时与每个病人只有几分钟的时间来检查症状，考虑可能的诊断，订购测试，并完成必要的文件。 这种工作量增加了出错的风险，给患者有意义的互动留下了较少的时间。 单靠雇用更多的工作人员不可能解决问题。
因此我认为员工持续短缺和工作量增加
医生可以信任AI什么？
最实际的应用之一是AI动力系统，记录医生和病人之间的交谈，总结会诊，并帮助编写临床文献。 医生不花时间手动写作指出，而是收到一份草稿，他们可以根据需要审查和编辑。
这些工具的好处在临床实践中已经很明显。 在JAMA发表的2026年4月的一项研究审查了来自美国五个医疗中心的8500多名保健专业人员的数据。 在采用人工智能文献工具的临床医生中，在指出临床上花费的时间平均减少了10%，相当于每8小时定期就诊约16分钟。
另一个例子是AI辅助的差别诊断。 想象一下，病人去看医生时，腹部疼痛而左侧有微弱的感觉。 AI系统可以分析症状，并提出几个可能的调查原因。 无法做最后的诊断，但可以帮助医生制定一份可能性清单，也希望它们成为医疗咨询的例行内容。 但是，它们也引起了一个合理的关注：依赖AI产生的建议是否会逐渐削弱医生独立理性的能力？当一个系统提供现成的答案时，它诱导人们相信它的结论，特别是当每个病人的时间已经很少的时候。
然而，还有另一面。 无论医生有多有经验，他们的知识不可避免地被他们治疗过的病人、他们接受的训练以及他们有时间研究所塑造。 与此同时，人工智能模型可以确定更多医疗数据的规律，包括个别医生可能从未遇到的病例。
大赦国际作为医学研究参与者
另一个值得探索的领域是复杂的临床病例，在这些病例中，甚至有经验的专家也挣扎着要达成共识。 如今，这类病例往往被提交医疗案例研究，由医生审查检测结果，讨论可能的诊断，并商定下一步的治疗步骤。 我认为AI不会取代这些讨论
有几种方法可以奏效。 多个模型可以独立地分析同一个临床病例，使医生能够比较其发现，并找出一致或有分歧的领域。 或者，单一模式可以使用不同的分析方法多次审查案件，然后将结果汇总起来审查。
当然，AI模型之间的一致并不一定意味着它们已经得出正确的结论。 模型可能共享限制，依赖类似的培训数据，并犯同样的错误。 或说明医生最初没有考虑过的可能性。
然而，大赦国际要对此类讨论作出有意义的贡献，就需要考虑的不仅仅是孤立的试验结果。 我亲眼看到这个更广泛的临床背景在我们Keymakr的作品中的重要性。 在一个超声波成像项目中，我们与不同时间拍摄的同一批病人的扫描工作，有时是一天几次，几个月后再次进行。 该团队对解剖结构作了注释，并标出潜在的异常情况，供专家审查，同时说明病人检查史的变化。
个别记录中小幅不一致可能影响整个数据集对有关数据的解释。 这一经验强化了将病人的病史作为一个整体考虑的重要性，这对于支持复杂临床决定的人工智能系统同样至关重要。
核磁共振没有核磁共振扫描仪
但诊断不仅仅是正确解释医疗数据。 也是为了获得这些数据。 在斯克里昂推荐的活动中，我遇到的最有趣的AI项目之一，一组开发人员已经找到了一种方法，可以使用CT扫描和从特定角度取取的多张X射线图像来利用人工智能来重建一个类似核磁共振的图像。 这个想法出自一个非常实际的问题：核磁共振扫描仪很昂贵，患者有时需要等几个月才能进行检查，而X射线机和CT扫描仪则能被广泛使用。
目前，技术侧重于膝盖，尽管开发者正在考虑将其扩展到其他关节。 团队认为，被重建的图像旨在提供与核磁共振扫描相似的信息，尽管成像技术依赖于不同的物理原理。
该团队不是开发另一台昂贵的医疗设备，而是利用现有的成像技术，并通过AI驱动的图像重建来扩展其所能做的。
无国界业务室
我期望大赦国际在今后十年中至少以这种或那种形式在所有医疗程序中发挥作用。 包括医疗数据分析和诊断支持
从长远来看，这些技术也有助于克服在保健方面的财政、组织甚至地理上的障碍。 想象一下，美国有技术的外科医生协助印度儿童做手术，或者印度有专家帮助治疗加拿大的病人，而不必在手术室里在场。 机器人、远程控制系统和人工智能辅助手术方面的进步可以使这种情况成为可能。
我们已看到朝这个方向迈出了第一步。 例如，Proximie平台允许外科医生远程连接手术室，实时观察手术程序，并与现场医疗队合作。 该系统支持直播视频，语音通信，和视觉说明，并使用AI分析外科工作流程数据。 这并不意味着外科医生可以远程进行手术
我认为，改善获得医疗保健的机会应该是医疗AI的中心目标之一。 重要的是这些技术是否能够帮助缩短诊断测试等候时间，减轻医疗专业人员的工作量，并给医生更多的时间与病人在一起。 最终，我将衡量医学AI的成功程度，衡量有多少人能够因此得到及时，高质量的医疗护理。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Thought Leaders
When AI Enters the Hospital: What Changes for Doctors and Patients
Add Unite.AI to your preferred sources on Google
In my previous column, I explored medical AI from a business perspective, focusing on the opportunities it creates for technology companies. But beyond the potential for innovation and new markets, there is a more fundamental question: what difference will these technologies make to the doctors who use them and the patients who depend on them? After all, the true value of medical innovation lies in its ability to improve the quality and accessibility of care. And I’m already seeing remarkable changes in how that care can be delivered.
How Staff Shortages Are Accelerating Healthcare Automation
I know an entrepreneur whose company provides temporary medical staff to hospitals, including nurses who fill in when regular employees are unavailable. You might expect a business like this to thrive amid persistent staffing shortages. But he says that even companies that specialize in supplying healthcare workers are struggling to find enough qualified people.
Hospitals have faced shortages of doctors and nurses for decades. Training new healthcare professionals takes considerable time and resources. Medical education, particularly in the United States, can be too expensive, and the years of study and clinical training come with financial and psychological pressure. Even after completing their training, doctors face high levels of stress, enormous responsibility, and constant time constraints.
As a result, physicians sometimes have just a few minutes with each patient to review symptoms, consider possible diagnoses, order tests, and complete the necessary documentation. This workload increases the risk of errors and leaves less time for meaningful patient interaction. And hiring more staff alone is unlikely to solve the problem.
That’s why I believe persistent staff shortages and growing workloads will be among the main forces accelerating AI adoption in hospitals.
What Can Doctors Entrust to AI?
One of the most practical applications is AI-powered systems that record conversations between doctors and patients, summarize consultations, and help prepare clinical documentation. Instead of spending time writing notes manually, physicians receive a draft they can review and edit as needed.
The benefits of these tools are already becoming evident in clinical practice. An April 2026 study published in JAMA examined data from more than 8,500 healthcare professionals across five US medical centers. Among clinicians who adopted AI-powered documentation tools, time spent on clinical notes decreased by an average of 10%, equivalent to approximately 16 minutes per eight-hour period of scheduled patient visits.
Another example is AI-assisted differential diagnosis. Imagine a patient visiting a doctor with abdominal pain and a tingling sensation on the left side. An AI system can analyze symptoms and suggest several possible causes to investigate. It doesn’t make the final diagnosis, but can help the physician develop a list of possibilities, identify follow-up questions, and determine which tests might be necessary. I’ve already encountered tools like these in practice, and I expect them to become a routine part of medical consultations. But they also raise a legitimate concern: could relying on AI-generated suggestions gradually weaken doctors’ ability to reason independently? When a system offers ready-made answers, it’s tempting to trust its conclusions, especially when there’s already so little time for each patient.
There is another side to this, however. No matter how experienced a physician is, their knowledge is inevitably shaped by the patients they’ve treated, the training they’ve received, and the research they’ve had time to study. AI models, meanwhile, can potentially identify patterns across much larger volumes of medical data, including cases an individual physician may never have encountered.
AI as a Participant in Medical Research
Another area worth exploring is complex clinical cases in which even experienced specialists struggle to reach consensus. Today, such cases are often brought before medical case studies, where physicians review test results, discuss possible diagnoses, and agree on the next steps in treatment. I don’t think AI will replace these discussions, but I can imagine it becoming another participant.
There are several ways this could work. Multiple models could independently analyze the same clinical case, allowing physicians to compare their findings and identify areas of agreement or disagreement. Alternatively, a single model could examine the case multiple times using different analytical approaches, with the results then brought together for review.
Of course, agreement among AI models doesn’t necessarily mean they’ve reached the correct conclusion. Models may share limitations, rely on similar training data, and make the same mistakes. Still, these additional perspectives could help uncover inconsistencies, flag rare conditions, or suggest possibilities that the physicians hadn’t initially considered.
For AI to contribute meaningfully to such discussions, however, it needs to consider more than isolated test results. I’ve seen the importance of this broader clinical context firsthand in our work at Keymakr. In one ultrasound imaging project, we worked with scans from the same patients taken at different times, sometimes several times a day and again months later. The team annotated anatomical structures and flagged potential abnormalities for specialist review, while accounting for changes across the patient’s examination history.
Small inconsistencies in individual records could affect the interpretation of related data across the entire span of the dataset. This experience reinforced how important it is to consider a patient’s medical history as a whole – something that would be equally essential for AI systems supporting complex clinical decisions.
MRI Without an MRI Scanner
But making a diagnosis isn’t just about interpreting medical data correctly. It’s also about being able to obtain that data in the first place. One of the most interesting AI projects I’ve come across at a Scryon-recommended event addresses exactly this challenge. A team of developers has found a way to use CT scans and several X-ray images taken from specific angles to reconstruct an MRI-like image using artificial intelligence. The idea emerged from a very practical problem: MRI scanners are expensive, and patients sometimes have to wait months for an examination, while X-ray machines and CT scanners are much more widely available.
For now, the technology focuses on the knee, although the developers are considering extending it to other joints. According to the team, the reconstructed images are intended to provide information comparable to MRI scans, even though the imaging techniques rely on different physical principles.
Rather than developing another expensive piece of medical equipment, the team is using existing imaging technology and expanding what it can do through AI-powered image reconstruction.
An Operating Room Without Borders
I expect AI to play a role in at least half of all medical processes in one form or another over the next decade. I’m talking about a broad range of applications, from medical data analysis and diagnostic support to real-time AI guidance during surgical procedures.
In the longer term, these technologies could also help overcome financial, organizational, and even geographical barriers to healthcare. Imagine a skilled surgeon in the United States assisting with an operation on a child in India, or a specialist in India helping treat a patient in Canada, without being physically present in the operating room. Advances in robotics, remote-control systems, and AI-powered surgical assistance could make scenarios like these possible.
We’re already seeing early steps in this direction. For example, the Proximie platform allows surgeons to connect remotely to operating rooms, observe procedures in real time, and collaborate with on-site medical teams. The system supports live video, voice communication, and visual annotations, and uses AI to analyze surgical workflow data. This doesn’t mean surgeons can already perform any operation remotely, but it does allow specialists to share their expertise and provide guidance without being in the same operating room.
I believe improving access to healthcare should be one of the central goals of medical AI. What matters is whether these technologies can help shorten diagnostic test wait times, ease the workload of medical professionals, and give doctors more time with their patients. Ultimately, I would measure the success of medical AI by how many more people can receive timely, high-quality medical care because of it.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.