编者按：由中国抗癌协会主办，天津医科大学肿瘤医院、天津市抗癌协会、中国整合医学发展战略研究院承办的“2023中国整合肿瘤学大会（2023 CCHIO）”将于2023年11月16—19日在天津举办。世界卫生组织国际癌症研究机构（WHO IACR）负责人Partha Basu教授将在会上就癌症筛查的AI发展和未来予以介绍。《肿瘤瞭望》就此开展特别采访，邀请Partha Basu教授分享全球癌症筛查面临的挑战、AI的发展以及中国的贡献。

 

Editor’s Note:Organized by the China Anti-Cancer Association and hosted by Tianjin Medical University Cancer Institute&Hospital，Tianjin Anti-Cancer Association，and the China Research Institute of Development Strategies of Holistic Integrative Medicine，the"2023 Chinese Congress of Holistic Integrative Oncology(2023 CCHIO)"will be held in Tianjin from November 16th to 19th，2023.Dr Partha Basu，the head of the Cancer Early Detection&Prevention Branch at the International Agency for Research on Cancer(IARC)of the World Health Organization(WHO)，will give the presentation of“Artificial intelligence and future of cancer screeningt”at the CCHIO conference.Oncology Frontier conducted a special interview to invite Dr Partha Basu to share the challenges of global cancer screening，the development of AI，and China’s contributions.

Partha Basu教授：世界卫生组织（WHO）开展了一项模型研究，以确定消除宫颈癌所需的时间。通过观察，我们确定，要在本世纪末实现该目标，就必须确保筛查的高覆盖率，至少要对70%的适龄女性进行一生两次的筛查，同时为15岁以下的女孩接种疫苗。此外，还要确保应用高质量的筛查，人乳头状瘤病毒（HPV）检测是当前准确性较高的筛查手段。另外，也须确保查出癌前病变和癌症的患者都能接受治疗。然而，当下的筛查、管理均存在诸多挑战。

 

首先是女性在获得筛查服务时遇到的阻碍。由于社会原因、经济原因以及对宫颈癌预防的认知不足，女性面临着许多诊疗阻碍。另一个挑战则是结构性阻碍或者卫生系统阻碍，即在中低收入国家暂无法实现女性筛查的高覆盖率、也无法为筛查阳性的患者提供合适的治疗。在许多资源有限的环境中，HPV检测仍然相当昂贵，而且无法获取。本着“即筛即治”（screen-and-treat）的原则，对部分患者可以采用热消融治疗，以实现较高的覆盖率或治疗依从性。另外约有30%~40%的患者需要接受切除手术治疗，如环电切术（LEEP术）等，但这也是一个巨大的挑战，因为切除性治疗需要在专业的治疗中心进行，但许多地方缺乏此类机构，导致生存率较低。总之，在确保妇女HPV筛查、治疗的高覆盖方面，目前仍在多个层面存在阻碍。

 

Dr.Partha Basu:The WHO conducted a modeling study to determine the milestones required to be achieved to eliminate cervical cancer.Through our observations，we determined that achieving the elimination target by the end of this millennium necessitates ensuring high screening coverage--at least twice in a lifetime screening for 70%of the age-eligible women along with vaccination of girls below 15 years of age.Additionally，it is crucial to guarantee the appropriate quality of the screening test.Currently，it can be stated with near certainty that the human papillomavirus(HPV)test is an exceptional test with high accuracy.We must also ensure that women who are diagnosed with pre-cancerous and cancerous lesions receive treatment.However，there are challenges in many aspects of the screening continuum.

 

Foremost are the barriers that a women encounters to access screening services.Women face numerous barriers due to social reasons，economic reasons，and their limited awareness of cervical cancer prevention.Therefore，there are barriers at the level of women.

 

There are even more significant barriers，which we refer to as structural barriers or health system barriers.In many countries，particularly low-and middle-income countries，we have yet to establish a system that can guarantee high coverage for women，ensure the provision of high-accuracy testing，and ensure appropriate management of positive screening results.HPV tests are still quite expensive and unavailable in many limited resourced settings.

 

By adopting a screen-and-treat approach，we can effectively treat many women through thermal ablation，resulting in a high compliance rate for treatment.However，approximately 30%to 40%of women require excisional treatment，also known as LEEP.This poses a significant challenge since these women need to be referred to specialized centers.In many areas，such facilities are not well-organized，leading to a high attrition rate in women seeking treatment.Therefore，achieving a high coverage of women and particularly offering them HPV detection tests involves several levels of barriers.As I mentioned earlier，we must address these barriers to ensure successful outcomes.

 

Partha Basu教授：人工智能（AI）在确保癌症筛查质量方面有很大的潜力。许多国家正在评估基于AI的癌症筛查，主要集中于乳腺癌、宫颈癌、结直肠癌、肺癌等。AI通过反复学习既往与最终诊断相关的诊疗数据辅助诊断，该原理在癌症筛查的各个方面都得到了应用。

 

首先，在宫颈细胞学方面，已有相关AI APP在部分国家获批用于解释宫颈细胞学，并在细胞学阅片方面表现良好。不容忽视的是，对于中低收入国家而言，阅片是难点，确保采集样本的质量、拥有高水平实验室采样技术人员同样是难点。如果样品载片质量差，AI也无法进行正常筛查，在临床实践中类似的挑战还有许多。

 

其次，AI可以分析乳腺钼靶以区分良性、恶性病变。在对数以千计的乳腺钼靶X光片进行训练学习后，已进行了广泛的研究，并开发出了高效的算法。欧洲专家建议在乳腺癌筛查时需安排第二阅片员对乳腺钼靶阴性的X光片进行复审。如今可以使用AI充当第二阅片员，基于现有循证医学证据，欧洲的指南也推荐了AI辅助筛查的方式。我们看到AI正在逐渐应用于癌症筛查，并在结肠镜检中筛查癌变非常有效。

 

第三，可以通过AI评估醋酸试验（VIA）后收集的图像，来辅助诊断宫颈癌前病变和宫颈癌。我们在宫颈癌筛查的重点是确保中低收入国家的妇女能够充分接受筛查和治疗，VIA是一种适用的方法。在中低收入国家，VIA筛查的应用与效果优于细胞学检查。VIA诊断的主要问题是主观性较强，需要对医务人员进行充分训练以确保诊断质量。利用同样的原理，可以训练AI算法来解读宫颈图像。目前有大量相关研究正在开展，包括国际癌症研究机构（IARC）利用AI识别VIA或碘试验检查图像以辅助诊断。在AI辅助检查的模式下，护士就可以通过拍摄VIA图像完成AI算法诊断。

 

此外，我们认为AI在HPV阳性患者分类方面有着更大的潜力。HPV检测的前景正在逐渐改变，全球有多种技术可用于HPV检测。随着时间的推移，负担能力将得到改善。在资源有限的情况下，HPV自取样提升了妇女的检测依从性。女性可以提供自采样本予以检测，其中阳性患者要进一步检查以确定具体病变情况，再考虑是否开展消融或切除治疗。这是AI的亮点所在。

 

因此，基于AI的宫颈图像分析或VIA后的外观分析是分流HPV阳性女性的一种非常可行且极具前景的方法，但仍需训练并建立完备的算法。过去两年间，我们使用泰国、印度等地上万影像训练了一个算法模型，目前IARC整在津巴布韦开展的一项大型临床研究对该算法予以验证，这是一个相当有前景的项目。不过我们意识到，要让AI获得高质量结果，高质量的图像的必不可少的。所以，简单的使用手机对宫颈拍照并用AI分析的想法并不可取。于是我们开发了一种图像捕捉设备以收集高质量的图像，在此基础上再行AI诊断。

 

AI领域非常有前景，但依然有许多尚未解答的问题。例如，受检女性以及医疗人员是否信任AI技术仍悬而未决；AI误诊等伦理和法律问题也有待商榷。这些都是我们未来需要解决的问题。

 

Dr.Partha Basu:Artificial intelligence(AI)holds a great potential in ensuring high-quality cancer screening.In many countries，AI-based cancer screening is being evaluated and mainly focuses on breast cancer，cervical cancer，colorectal cancer，and lung cancer.AI examines the pattern or repetitive data linked to‘ground truth’，the final diagnosis，and tries to interpret it to produce a diagnosis as a result.This principle has been applied in cancer screening in different aspects.

 

Firstly，in cervical cytology，there are already validated AI-based applications，which are approved in some countries for interpreting cervical cytology.These applications have been found to perform well in cytology reading.However，when discussing cytology，it is important to recognize that the challenge is not just in reading the cytology，especially for low-and middle-income countries.Ensuring good quality sample collection and having available well-trained laboratory technicians who can produce high-quality slides is crucial.If the slide quality is poor，even the best AI technology will not work effectively.This is an example where AI has been developed and is functioning，but there are several gaps in practical applications.

 

Secondly，AI can examine the patterns visible in mammography films to identify malignant lesions，differentiate between benign and malignant lesions，or identify truly normal mammograms.Extensive studies have been conducted and highly effective algorithms have been developed after being trained on thousands of mammography images.In many European countries each negative mammogram should be reviewed by a second reader.We can use an AI-based algorithm to read negative mammograms instead of a second reader.This approach is supported by strong evidence and has already been recommended in European guidelines.You can see that AI is gradually being incorporated into cancer screening.Similarly，for colorectal cancer screening，AI has been found to be highly effective in detecting lesions during colonoscopy.

 

Thirdly，AI can be employed to assist in the diagnosis of cervical precancers and cancers through evaluation of images collected after application of acetic acid(VIA).In cervical cancer screening，our primary focus is on ensuring the screening and management of women in low-and middle-income countries，where several different opportunities arise.One such opportunity is visual inspection with VIA，which has shown great promise.At least it is much better than doing cytology in a low-and middle-income country setting.However，the main challenge of VIA is that it is subjective and requires extensive training of the providers to perform high-quality VIA assessments.Using the same principles AI algorithms can be trained to interpret cervical images.Extensive work is ongoing，including an International Agency for Research on Cancer study that is exploring AI-based systems to recognize patterns visible in cervical images after the application of acetic acid or Lugol’s solution.In a screening setting，women are examined by nurses who apply acetic acid to the cervix，capture the image，and then an AI-based algorithm provides the diagnosis.

 

Furthermore，we believe that AI holds greater potential in the triage of HPV-positive women.As you are aware，the landscape of HPV testing is gradually shifting.There are several technologies now available worldwide.Affordability will improve over time.HPV self-sampling has created a significant opportunity to improve compliance among women，particularly in limited-resource settings.Women can now opt for HPV testing and provide self-collected samples，but positive cases require examination through triage to identify those with lesions and those without lesions.Additionally，decisions need to be made on whether they should undergo ablation or excisional treatment.This is where AI truly shines.

 

Therefore，an AI-based analysis of cervical images or the appearance after VIA is a very possible approach for triaging HPV positive women，which is a very promising aspect，but we still need to have appropriately trained algorithms.We need to validate the algorithm in a field setting，and that is exactly what a large IARC trial is doing in Zimbabwe.So we have developed an algorithm and trained the algorithm over the last two years using thousands of images collected from Thailand，India，and other countries.This algorithm will be field-tested in Zimbabwe.So this is，we think，a very promising project.We have come to the realization that for AI development to achieve good quality results，good quality images are necessary.Therefore，the original concept of simply using a mobile phone to capture an image of the cervix for AI analysis may not be entirely practical.As such，we are also developing an image-capturing device that will ensure high-quality image capture，which can then be used by the AI for diagnosis.

 

AI is a highly promising field，but there are still many unanswered questions.For instance，the trust of women and those being screened in the AI system remains an unresolved issue.Similarly，the trust of healthcare professionals in AI has yet to be addressed.There are also several ethical and legal concerns，such as who would be responsible if an AI-based algorithm makes a mistake.These are matters that we still need to resolve.

Partha Basu教授：我们的中国同行为我们了解各种检测、分诊和治疗技术的准确性做出了巨大贡献，特别是在社区环境中宫颈癌前病变等的筛查方面。中国在农村和城市开展的大规模筛查项目中嵌入了多项研究，为乳腺癌、宫颈癌、胃癌、食管癌等癌症的筛查提供了非常有价值的证据。CACA营造的全球合作氛围将有助于国际抗肿瘤组织交流思想、分享信息，并指导全球抗肿瘤策略发展。中国同行所开展的多项应研究为2021年、2022年WHO发表的宫颈癌筛查和管理指南做出了贡献。在AI方面，同样有大量来自中国的研究涵盖了癌症筛查的方方面面。由于AI的人种特异性，可能基于中国人群和影像开发的AI并不适合非洲人群，反之亦然。因此，在AI的开发过程中，需要更多的全球合作来消除人种偏倚，以确保AI系统可以在中国人群、其他亚洲人群、非洲人群或拉丁美洲人群中同样适用。

 

中国对改善全球癌症筛查的另一项重要贡献是引入了更经济实惠、且经过充分验证的检测方案，其中许多检测方案已扩展到许多国家。由于规模优势，中国开发的检测耗材和技术成本更加经济实惠。因此，我坚定地认为，CACA倡议在癌症早期检测和预防等方面建立一个合作程度更高的研究环境，是值得高度赞扬的。

 

Dr.Partha Basu:Our colleagues in China have made immense contributions to our understanding of the accuracy of various testing and triaging technologies，including treatment methods，particularly for cervical precancers in community settings.The screening programs that exist in China in rural and urban areas，are large-scale programs，and several studies have been nested within these programs to generate extremely valuable evidence not only for breast and cervical cancer screenings，but also for some of the uncommon screening sites such as gastric and esophageal cancers，as well as liver cancer.Therefore，I believe that the collaborative atmosphere fostered by CACA will assist the global community in terms of exchanging ideas，sharing information，and guiding global policies.Numerous studies conducted by our colleagues in China have significantly informed the WHO guidelines for cervical cancer screening and management published in 2021 and 2022.Additionally，in the realm of AI，China has published numerous studies examining AI-based systems in different aspects of cancer screening continuum.

 

It is essential to collaborate with the global community，as we recognize that AI has a population-specific element.This implies that an AI developed using data and images from the Chinese population may not perform equally well when applied to populations in Africa，and vice versa.Therefore，greater collaboration is necessary when it comes to the development of AI across diverse global communities.This ensures trustworthy systems that can perform equally well in Chinese，Asian，African，or Latin American populations.It is crucial to eliminate any bias in AI development to promote inclusivity and enhance the accuracy of these systems.

 

Another significant contribution of China to global efforts in improving cancer screening is by introducing more affordable testing options，which have been thoroughly validated in China.Many of these tests are already being utilized in China and beyond.Due to the vast scale of production，the costs of consumables and technologies developed in China are more affordable compared to similar technologies developed elsewhere in the world.Hence，I firmly believe that CACA’s initiative to foster a more collaborative research environment，particularly in cancer early detection and prevention，is highly commendable.