剑桥研究人员正在探索人工智能如何在药物研发、阿尔茨海默症诊断以及全科医生问诊等多个领域实现变革。

Cambridge researchers are looking at ways that AI can transform everything from drug discovery to Alzheimer's diagnoses to GP consultations.

2024年，心理学系的佐伊·库尔齐（Professor Zoe Kourtzi）教授展示了一款由其团队开发的人工智能工具，相较于临床测试，该工具在预测已有早期痴呆症症状的患者的病情，是会保持稳定或继续发展为阿尔茨海默病方面表现更优。

In 2024, Professor Zoe Kourtzi in the Department of Psychology showed that an AI tool developed by her team could outperform clinical tests at predicting whether people with early signs of dementia will remain stable or develop Alzheimer’s disease.

在英国国家医疗服务体系（NHS）面临巨大压力的当下，这类工具可以帮助医生优先照顾最需要治疗的患者，同时也能避免对病情稳定的患者进行侵入性和昂贵的诊断测试。这类工具也能让病人因知道他们的病情恶化可能性不大而安心；而对不那么幸运的患者，这类工具则可以帮助他们及家人早做准备。

At a time of intense pressure on the NHS, tools such as this could help doctors prioritise care for those patients who need it most, while removing the need for invasive and costly diagnostic tests for those whose condition will remain stable. They can also give patients peace of mind that their condition is unlikely to worsen, or, for those less fortunate, it can help them and their families prepare.

库尔齐表示，这些工具还可能在新药研发中发挥变革性作用，使临床试验更高效、更快捷、更低成本。

These tools could also be transformational in the search for new drugs, making clinical trials more effective, faster and cheaper, says Kourtzi.

近期，治疗痴呆的两种药物——仑卡奈单抗（lecanemab）和多奈单抗（donanemab）——在延缓病程方面表现出一定潜力，但其效益与成本之比被认为不足以支持在NHS体系内的批准使用。除此之外，该领域的进展仍较有限。

Recently, two dementia drugs – lecanemab and donanemab – have shown promise in slowing the disease, but the benefits compared to the costs were judged insufficient to warrant approval for use within the NHS. Beyond these, there’s been limited progress in the field.

部分问题在于临床试验常常选错了受试人群，而人工智能可以协助更好判断应让哪些人参与试验。

Part of the problem is that clinical trials often focus on the wrong people, which is where AI may help to better decide who to include in trials.

“如果人工智能模型表明某些人不会出现病理变化，那你就不应该让他们参与试验，他们只会干扰统计结果。哪怕你拥有全球最好的药物，试验也不会显示出实际效果。而如果你纳入的是病情发展过快的人，药物可能已经来不及发挥作用了。”

“If you have people that the AI models say will not develop pathology, you won't want to put them in your trial. They'll only mess up the statistics, and then [the trials] will never show an effect, no matter if you have the best drug in the world. And if you include people who will progress really fast, it might be already too late for the drug to show benefit.”

库尔齐正领导ai@cam的一个“AI创意（AI-deas）”项目，旨在建立“脑健康中心（BrainHealth hub）”，以应对全球大脑与心理健康危机。

Kourtzi is leading one of ai@cam’s AI-deas projects to create a ‘BrainHealth hub’ to tackle the global brain and mental health crisis.

该项目可以弥合两个群体之间的鸿沟：一类是拥有技术但缺乏数据的工程师、数学家和计算机科学家；另一类是拥有数据却缺乏高级工具进行挖掘分析的临床医生和神经科学家。

“我们的想法是创建一个‘创意温室’，让人们聚在一起提出并解答各种富有挑战性的问题。”

“Our idea is to create a ‘hothouse’ of ideas where people can come together to ask and answer challenging questions.“

大学研究人员、产业合作伙伴、慈善组织和政策制定者可以共同探讨诸多问题，如我们如何利用人工智能进行新药发现、加速临床试验并开发新疗法？我们如何构建可解释的人工智能模型，使其转化为临床工具？

University researchers, industry partners, the charity sector and policymakers will explore questions such as: how can we use AI for drug discovery, to accelerate clinical trials and develop new treatments, and how can we build interpretable AI models that can be translated to clinical tools?”

每当库尔齐与患者群体交流时，反复被提到的话题就是：这类人工智能必须是可靠且可被信任的。

The need for such AI to be reliable and responsible is a theme that comes up frequently when Kourtzi speaks to patient groups.

“当医生使用像核磁共振这样的复杂诊断工具时，患者并不会质疑他们是否理解机器的原理或运行方式。他们关心的是：这设备是否通过了监管标准、是否安全可靠。对人工智能而言，情况也是一样的。”

“When doctors are using a complex diagnostic tool like an MRI machine, patients don't query whether they understand what's in this machine, why it works this way. What they want to know is that it's gone through regulatory standards, it's safe to use and can be trusted. It’s exactly the same with AI.”

来自“医疗改善研究所”（THIS）的尼尔斯·皮克教授（Professor Niels Peek）认为，人工智能或能通过处理一些最为繁琐的工作对如全科医生工作等的基础保健服务产生重大影响。

Professor Niels Peek from The Healthcare Improvement Studies (THIS) Institute believes that AI could have a major impact on primary care services, such as GP practices, by tackling some of their most mundane tasks.

其中一个应用是使用“数字书记员”来记录、转录和总结全科医生与患者之间的对话。“你去算算临床医生在这些工作上花费的时间，那真是令人震惊。”他说。

One such application involves the use of ‘digital scribes’ to record, transcribe, and summarise conversations between GPs and patients. “If you look at the amount of time that clinicians spend on that type of work, it’s just incredible,” he says.

“考虑到临床医生的时间可能是NHS中最宝贵的资源，这是能够造成变革性影响的技术。”

“Considering that clinician time is probably the most precious commodity within the NHS, this is technology that could be transformational.”

未来，NHS很可能会越来越多地采用数字化速记技术，因此确保这些总结准确无误、不遗漏关键点或加入没有提到的内容（即“幻听”）非常重要。在健康基金会的支持下，皮克正在研究这一技术是否真的能节省时间。“如果你需要花很多时间去修正它的输出，那么它不再是一个节省时间的工具了。”

It is likely that the NHS will increasingly adopt digital scribe technology in the future, so it will be important to ensure the summaries are accurate and do not omit key points or add things that were not mentioned (a ‘hallucination’). With support from The Health Foundation, Peek is asking whether the technology actually saves time? “If you have to spend a lot of time correcting its outputs, then it's no longer a given that it actually does save you time.”

皮克认为，未来每次临床问诊都将被数字化记录下来、存储为患者记录的一部分，并由人工智能进行总结。但鉴于现有的技术环境，特别是在基础保健领域，这是个不小的挑战。

Peek believes that in the future, every clinical consultation will be recorded digitally, stored as part of a patient's record, and summarised with AI. But the existing technology environment, particularly in primary care, presents a challenge.

“全科医生使用的电子健康记录系统经历了多年的发展，往往显得过时。任何新技术都必须能够适应这些系统中。让人们登录到不同的系统来完成工作是不可行的。”

“GPs use electronic health records that have evolved over time and often look outdated. Any new technology must fit within these systems. Asking people to log into a different system is not feasible.”

皮克还参与了“Patchs”工具的评估，该工具将人工智能应用于预约全科医生并进行在线问诊的流程。Patchs由全科诊所的工作人员、患者与曼彻斯特大学（皮克曾在该校工作）合作设计，并由Patchs Health公司进行商业化。目前，英格兰约有十分之一的全科诊所使用这一工具。

Peek is also involved in evaluating Patchs, a tool that applies AI to the process of booking GP appointments and conducting online consultations. It was designed by GP staff and patients, in collaboration with The University of Manchester (where Peek was formerly based) and commercialised by the company Patchs Health. It is now used by around one in 10 GP practices across England.

与终端用户，也就是患者、全科医生，特别是日常使用这些系统的行政人员进行合作至关重要。“你必须确保它们不仅与人们已经使用的系统兼容，而且与他们的工作流程相符。只有这样，你才能看到这些变化是实实在在地为人们带来便利。”

Working with end users – patients, GPs, and particularly the administrative staff who use these systems on a day-to-day basis – is crucial.  “You have to make sure they fit both with the systems people are already using, and also with how they do things, with their workflows. Only then will you see differences that translate into real benefits to people.”

近年来，年轻人中心理健康障碍的患病率显著增加。然而，由于英国国家卫生服务（NHS）资源紧张，年轻人往往难以获得儿童与青少年心理健康服务（CAMHS）。

Over recent years, there has been a significant increase in the prevalence of mental health disorders among young people. But with stretched NHS resources, it can be difficult to access Child and Adolescent Mental Health Services (CAMHS).

精神病学系的安娜·摩尔博士（Dr Anna Moore）表示，并非每个被推荐转诊的孩子都需要看心理疾病专科医生，但问题是他们可能需要等上两年才能被告知他们不符合治疗标准。而且，他们得到的符合其需求的替代方案建议的质量差异很大。

Not every child recommended for a referral will need to see a mental health specialist, says Dr Anna Moore from the Department of Psychiatry, but the bottleneck means they can be on the waiting list for up to two years only to be told they don’t meet the criteria for treatment. The quality of advice they get about alternative options that do meet their needs varies a lot.

她说：“帮助我们实现这一目标的数据可能是一些非常敏感的个人信息。像是健康信息、孩子在学校的表现，也可能是像上周末喝醉并最终去急诊的情况。”

“The kinds of data that help us do this can be some of the really sensitive data about people,” she says. “It might be health information, how they're doing at school, but it could also be information such as they got drunk last weekend and ended up in A&E.”

因此，她表示，在设计这种系统时，必须与公众紧密合作，以确保人们理解他们的工作内容、正考虑使用的数据种类及其用途，同时也要理解这一系统如何有可能改善对有心理健康问题的年轻人的护理。

For this reason, she says, it’s essential that they work closely with members of the public when designing such a system to make sure people understand what they are doing, the kinds of data they are considering using and how it might be used, but also how it might improve the care of young people with mental health problems.

伦理学家经常提出的一个问题是，考虑到接触到CAMHS服务的困难，如果无法让孩子们获得服务，找到他们真的是件好事吗？

One of the questions that often comes up from ethicists is whether, given the difficulties in accessing CAMHS, it is necessarily a good thing to identify children if they cannot then access services.

她表示：“是的，我们可以识别出那些需要帮助的孩子，但我们需要问一个问题，‘然后呢？’”该工具得将需要的孩子推荐转诊到CAMHS，但对于那些有问题但可以通过比CAMHS更灵活的其他方式获得更好支持的孩子，它能否将他们指引到有帮助的、基于证据的、适龄的信息？

“Yes, we can identify those kids who need help, but we need to ask, ‘but so what?’,” she says. The tool will need to suggest a referral to CAMHS for the children who need it, but for those who have a problem but could be better supported in other ways than CAMHS that could be more flexible to their needs, can it signpost them to helpful, evidence-based, age-appropriate information?

为了帮助找到那些可能被忽视的孩子，摩尔博士正致力于设计这款工具。在最极端的情况下，这些孩子可能是像维多利亚·克林比（Victoria Climbié）和婴儿P（Baby P）这样的孩子，他们被监护人虐待并致死。对重大案件的回顾显示出，多次错失采取行动的机会往往是因为系统没有有效衔接，导致没有人能够看到全貌。

Moore is designing the tool to help find those children who might otherwise get missed. In the most extreme cases, these might be children such as Victoria Climbié and Baby P, who were tortured and murdered by their guardians. The serious case reviews highlighted multiple missed opportunities for action, often because systems were not joined up, meaning no one was able to see full picture.

她说：“如果我们能够查看整个系统中与孩子相关的所有数据，那么就有可能将这些数据整合在一起，也就是说，实际上这里有足够的信息让我们能够采取行动。”

“If we're able to look at all of the data across the system relating to a child, then it might well be possible to bring that together and say, actually there's enough information here that we can do something about it.”

全球生育率正在下降，越来越多的家庭选择晚育。为了帮助他们怀孕，许多夫妻转向辅助生殖技术，如试管受精（IVF）；然而，成功率依然较低，而且过程费用高昂。在英国，私立诊所的诊疗费用每周期超过5000英镑，在美国则大约为2万美元，而且并没有成功的保证。

Across the world, fertility rates are falling, while families are choosing to have children later on in life. To help them conceive, many couples turn to assisted reproductive technologies such as IVF; however, success rates remain low and the process can be expensive. In the UK, treatment at a private clinic can cost more than £5,000 per cycle – in the US, it can be around $20,000 – and with no guarantee of success.

莫·瓦利（Mo Vali）和斯塔西·韦斯（Dr Staci Weiss）希望人工智能能够改变这一局面。他们领导着“从子宫到世界”项目，这是ai@cam的旗舰AI-deas项目之一，旨在通过早期诊断生育问题并个性化生育治疗，提高准父母怀孕的机会。

Mo Vali and Dr Staci Weiss hope that AI can change this. They are leading From Womb to World, one of ai@cam’s flagship AI-deas projects, which aims to improve prospective parents’ chances of having a baby by diagnosing fertility conditions early on and personalising fertility treatments.

目前，他们正与英国最大的独立私立试管婴儿诊所之一——利斯特生育诊所的汤姆·姚教授合作，旨在开发更便宜、更少侵入性且更准确的人工智能辅助测试，贯穿患者的整个试管受精过程。为此，他们正在利用生育过程中收集的各种不同样本和数据集，从血液测试和超声图像到卵泡液，以及包括人口统计和文化因素在内的数据。

They are working with Professor Yau Thum at The Lister Fertility Clinic, one of the largest standalone private IVF clinics in the UK, to develop cheaper, less invasive and more accurate AI-assisted tests that can be used throughout the patient’s IVF journey. To do this, they are making use of the myriad different samples and datasets collected during the fertility process, from blood tests and ultrasound images to follicular fluid, as well as data encompassing demographic and cultural factors.

团队还希望利用人工智能改进并使普及4D超声扫描，这种扫描可以让父母看到胎儿在子宫中的运动，捕捉到如吮拇指和打哈欠等动作。韦斯表示，这对于在潜在的压力时期加强母性纽带很重要。

The team also hopes to use AI to improve, and make more accessible, 4D ultrasound scans that let the parents see their baby moving in the womb, capturing movements like thumb-sucking and yawning. This is important for strengthening the maternal bond during a potentially stressful time, says Weiss.

“看着婴儿的脸，看着他/她的动作，会产生一种非常不同的生理反应，并加深母亲与孩子之间的纽带，”她说。

“Seeing their baby's face and watching it move creates a very different kind of physical, embodied reality and a bond between the mother and her child,” she says.

与经历过生育治疗挑战的女性开展咨询这一程序提供了宝贵的见解，而利斯特生育诊所作为一家私立诊所，是一个在将工具提供给更广泛的公众之前用来测试他们创意的理想平台。该诊所提供了一个更小、更受控的环境，他们可以直接与资深临床医生互动。

Consulting with women who have experienced first-hand the challenges of fertility treatments is providing valuable insights, while The Lister Fertility Clinic – a private clinic – is an ideal platform in which to test their ideas before providing tools for the wider public. It offers a smaller, more controlled environment where they can engage directly with senior clinicians.

“我们希望确保我们所做的研究和我们所构建的人工智能模型能够无缝工作，之后再进行大规模推广，”瓦利说。

“We want to ensure that the research that we are doing and the AI models that we're building work seamlessly before we go at scale,” says Vali.

剑桥大学癌症风险预测教授安东尼斯·安东尼乌（Antonis Antoniou）将大部分职业生涯都致力于开发预测人类患癌风险的模型。而如今，人工智能有望将他的研究提升到全新的高度。

Antonis Antoniou, Professor of Cancer Risk Prediction at Cambridge, has spent most of his career developing models that predict our risk of developing cancers. Now, AI promises to take his work to an entirely new level.

近日，安东尼乌被任命为“癌症数据驱动发现计划”（Cancer Data-Driven Discovery Programme）主任。这项耗资1000万英镑的计划旨在彻底变革我们未来检测、诊断乃至预防癌症的方式。这是一个合作伙伴遍布英国各地的跨机构的项目，该项目将建立基础设施并创建一个多学科研究团队，包括培养下一代研究人员，为癌症数据科学领域的30个博士名额和早期职业研究职位提供资金。

Antoniou has recently been announced as Director of the Cancer Data-Driven Discovery Programme, a £10million initiative that promises to transform how we detect, diagnose – and even prevent – cancer in the future. It’s a multi-institutional project, with partners across the UK, that will build infrastructure and create a multidisciplinary community of researchers, including training the next generation of researchers, with funding for 30 PhD places and early career research positions in cancer data sciences.

该项目将使科学家能够访问并整合各种多样化的健康数据集，从全科医生诊所和癌症筛查项目，到大型队列研究，甚至包括来自公共服务互动中产生的数据，例如职业、教育程度，以及关于空气污染、住房质量和公共服务可达性的地理空间数据。这些数据将与人工智能和最先进的分析技术相结合使用。

The programme will enable scientists to access and link a vast array of diverse health data sets, from GP clinics and cancer screening programmes to large cohort studies through to data generated through interactions with public services such as on occupation, educational attainment and other geospatial data on air pollution, housing quality and access to services. These will be used in combination with AI and state-of-the-art analytics.

“这笔资助将使我们能够利用这些数据集开发模型来预测个体患癌风险，并大幅度提高我们对高风险人群的识别能力，”他说，“希望这将有助于我们彻底改变未来癌症的检测、预防和诊断方式。”

“The funding will allow us to use these data sets to develop models that help us predict individual cancer risk and greatly improve our understanding of who is most at risk of developing cancer,” he says. “It will hopefully help us transform how we detect and prevent and diagnose cancer in the future.”

该团队研究的一个关键考虑因素是确保他们开发的人工智能工具不会无意中加剧社会不平等。

One of the key considerations of their work will be to ensure that the AI tools they develop do not inadvertently exacerbate inequalities.

“我们必须警惕不要开发出只对那些愿意参与研究或经常与医疗系统互动的人有效的模型，也要确保我们没有忽视那些难以获得医疗服务的人，比如住在贫困地区的人群。”

“We have to be careful not to develop models that only work for people who are willing to participate in research studies or those who frequently interact with the healthcare sector, for example, and ensure we’re not ignoring those who can’t easily access healthcare services, perhaps because they live in areas of deprivation.”

患者和公众的参与是该项目的关键。他们与临床医生一道，从项目初期就协助构建整体方案。

Key to their programme has been the involvement of patients and members of the public, who, alongside clinical practitioners, have helped them from the outset to shape their programme.

“他们从项目规划阶段就参与到方案的共同制定中，未来他们仍将发挥关键作用，指导我们的工作方式，并确保数据被负责任、安全地使用。”他说。

“They were involved in co-developing our proposals from the planning phase, and going forward, they’ll continue to play a key role, helping guide how we work and to make sure that the data are used responsibly and safely,” he says.

“癌症数据驱动发现计划”由英国癌症研究会（Cancer Research UK）、国家卫生与护理研究院（NIHR）、工程与自然科学研究委员会（EPSRC）、英国健康数据研究机构（Health Data Research UK）以及英国行政数据研究机构（Administrative Data Research UK）共同支持。

The Cancer Data-Driven Detection programme is jointly supported by Cancer Research UK, the National Institute for Health & Care Research, the Engineering & Physical Sciences Research Council, Health Data Research UK, and Administrative Data Research UK.

自首次完成人体基因组测序以来，仅仅二十多年，基因组学这一全新的科学领域应运而生，促进了我们对人体运作机制的理解。此后，诸如蛋白质组学（proteomics）、代谢组学（metabolomics）等以全面解析特定类型生物分子的“组学”数量激增。

It’s just over 20 years since the first human genome was sequenced, opening up a new scientific field – genomics – and helping us understand how our bodies function. Since then, the number of so-called ‘omics’ – complete readouts of particular types of molecules in our bodies, such as proteins (proteomics) and metabolites (metabolomics) – has blossomed.

米尔纳治疗研究所（Milner Therapeutics Institute）的Namshik Han博士致力于探索人工智能如何挖掘这些“组学”宝藏，从而推动新药发现。

Dr Namshik Han from the Milner Therapeutics Institute is interested in how AI can mine this treasure trove to help discover new drugs.

米尔纳研究所充当着学术界与产业界之间的桥梁，通过与数十家高校、行业机构、生物技术公司、制药企业和风险投资人合作，来加速研究的流程。而Han博士的工作“前沿”是与高科技企业的合作。

The Milner Institute acts as a bridge between academia and industry to accelerate this process, partnering with dozens of academic institutes, industry partners, biotech, pharma and venture capitalists. But at the ‘bleeding edge’ of Han’s work is his collaborations with tech companies.

Han感兴趣的是如何利用量子力学原理实现更快、更强大的计算的量子计算机，可以解决诸如支持药物开发的复杂化学等问题。

Han is interested in how quantum computers, which use principles of quantum mechanics to enable much faster and more powerful calculations, can address problems such as the complex chemistry underpinning drug development.

“我们已经证明量子算法能发现传统人工智能算法无法察觉的模式。”Han博士表示。

“We’ve shown that quantum algorithms see things that conventional AI algorithms don’t.” Han says.

他的实验室运用量子算法分析由上万种人类蛋白质构成的巨大网络。传统人工智能往往只能探索网络的部分区域，而Han博士的研究显示，量子算法可以全面覆盖整个网络结构。

His lab has used quantum algorithm to explore massive networks comprising tens of thousands of human proteins. When conventional AI explores these networks, it only looks at certain areas, whereas Han showed that quantum algorithms cover the entire network.

人工智能有望全面提升药物研发的各个环节——从靶点发现（如Han博士正在做的项目），到优化临床试验，从而降低新药开发成本，并让患者更快受益。但这还不是Han博士最为期待的突破。

AI has the potential to improve every aspect of drug discovery – from identifying targets, as Han is doing, to optimising clinical trials, potentially reducing the cost of new medications and ensuring patients benefit faster. But that’s not what really excites Han.

“以癌症为例，”他说，“癌症种类繁多，其中很多我们并没有专门的药物可治，只能借用治疗相似癌症的药品，而这是远非理想的方式。”

“Take cancer, for example,” he says. “There are many different types, and for some of them we don’t have specific drugs to treat them. Instead, we have to use a drug for a related cancer and give that to the patient, which is not ideal.

“基于量子的人工智能将为我们打开一扇全新的大门，让我们找到前所未有、真正具有创新性的药物。这才是真正影响力的所在。”

“Quantum based AI will open up a completely new door to find truly innovative drugs which we’ve never thought of before. That’s where the real impact has to be.”