ColonSys: an AI-powered risk stratification tool for improving FIT-based bowel cancer screening diagnostic yield

** 需求：** 英国的肠癌筛查计划（BCSP）已经通过改进诊断方法，粪便免疫化学测试（FIT）进行了革命性的改革。FIT很容易进行（只需要一个样本），因此50岁以上合格成年人的筛查参与率从59.4%增加到67.5%。任何超过预先确定的FIT阈值（普遍适用于所有筛查样本）的参与者都将接受结肠镜检查，以确定任何医学上可行的出血原因，即，结肠直肠癌（CRC）或癌前腺瘤（息肉）。与传统的基于愈创木脂的粪便血红蛋白（f-Hb）检测相比，FIT的灵敏度增加，结肠镜检查的转诊数量正在增加。约45%接受结肠镜检查的受试者既没有结直肠癌也没有腺瘤。（在FIT筛查轮之间检测到的癌症）占诊断的肠癌的约50%。间隔期癌症的高比例可以通过高阳性f-Hb FIT阈值（英格兰120 µg/g：苏格兰80 µg/g）来解释。最近的模型表明，NHS英格兰FIT阈值为120 µg/g可能意味着超过52%的CRC和75%的高风险腺瘤可能会被遗漏。英国肠癌和领先的临床医生建议FIT阈值应降低至20 µg/g，以促进CRC的早期检测，但这可能会增加结肠镜检查的数量，从650，结肠镜检查是一种昂贵的、侵入性的程序（每次结肠镜检查约550英镑）。此外，在2019冠状病毒病大流行期间，减少不必要的程序和工作人员/患者医院互动以控制感染的额外动力，同时通过更明智地使用关键资源来提高NHS的效率。因此，提高FIT的诊断率和避免不必要的结肠镜检查是一个重大而紧迫的 ** 未满足的需求 **。解决方案：** Advanced Expert Systems Ltd（AES）与来自邓迪大学和考文垂大学医院和沃里克郡米德兰NHS信托的临床学者合作，开发了一个基于人工智能的软件平台“ColonSys”，用于在其他风险因素的背景下解释FIT结果。作为一个多变量分层模型，ColonSys根据风险优化临床资源，从高风险患者快速跟踪以进行进一步调查，到无风险患者推迟下一次FIT筛查。ColonSys将使BCSP更有效，只有高风险患者需要结肠镜检查。**及时性：** 由于新冠肺炎，减少不必要的医院就诊和提高NHS效率的动力增加。

**NEED:** The UK's Bowel Cancer Screening Programme (BCSP) has been revolutionised by improved diagnostic methodology, the Faecal Immunochemical Test (FIT). FIT is easy to undertake (only one sample required), and so screening participation has increased from 59.4% to 67.5% of eligible adults aged \>50 years.Any participant exceeding a pre-determined FIT threshold value that is universally applied to all samples screened is referred for colonoscopy to identify any medically actionable cause of bleeding i.e., colorectal cancer (CRC) or a precancerous adenoma (polyp).Coupled to increased sensitivity of FIT vs. traditional guaiac-based faecal haemoglobin (f-Hb) detection, the number of referrals to colonoscopy is rising.There are recognised weaknesses with FIT; ~45% of participants undergoing colonoscopy have neither CRC nor adenomas.More worryingly however is that interval cancers (cancers that are detected between FIT screening rounds) account for ~50% of diagnosed bowel cancers. The high proportion of interval cancers can be explained by the high positive f-Hb FIT threshold (120 µg/g in England: 80 µg/g in Scotland). Recent modelling suggests that the NHS England FIT threshold of 120 µg/g could mean over 52% of CRC and 75% of high-risk adenomas may be missed.Bowel Cancer UK and leading clinicians have recommended that the FIT threshold should be lowered to 20 µg/g to facilitate earlier detection of CRC, but this could increase the number of colonoscopies being performed from 650,000 to as many as 1.5 million procedures annually.Colonoscopy is a costly, invasive procedure (c.£550 per colonoscopy). Moreover, during the Covid-19 pandemic, there is additional impetus to reduce unnecessary procedures and staff/patient hospital interactions for infection control, whilst making the NHS more efficient by using critical resources more wisely.There is thus a significant and pressing **unmet need** for improving the diagnostic yield from FIT and avoiding unnecessary colonoscopy.**SOLUTION:** Advanced Expert Systems Ltd (AES) have joined clinical academics from the University of Dundee and University Hospitals Coventry & Warwickshire Midlands NHS Trust to develop an AI-based software platform 'ColonSys' to interpret FIT result in the context of other risk factors.As a multivariate stratification model, ColonSys optimises clinical resources by risk, with a gradient from high-risk patients being fast-tracked for further investigation, to patients with no risk having their next FIT screen delayed.ColonSys will make BCSPs more efficient, with only high-risk patients requiring colonoscopy.**TIMELINESS:** Due to Covid-19, there is additional impetus to reduce unrequired hospital visits and improve NHS efficiency.