Establishing which modalities of artificial intelligence/machine learning (AI/ML) for early detection and diagnosis of cancer are ready for implementation in primary care in the NHS: a Systematic Review

Start Date Jan 2019

Code C11-C

Status Ongoing

Project Lead
Senior Lead
Dr Natalia Calanzani, Dr Evie Papavasiliou, Dr Valerie Sills, Prof M van der Schaar (all Cambridge), Prof Stephen Duffy (Queen Mary University of London), Prof Jon Emery (Melbourne), Dr RN Matin (Oxford), Prof Willie Hamilton (Exeter), Prof HC Williams (Nottingham), Prof Hardeep Singh (Baylor), Prof Niek de Wit (Utrecht),


Recently there has been huge interest in the application of artificial intelligence (AI) and machine learning (ML) to medicine, and there is accumulating evidence that AI can assist clinicians to make better clinical decisions or even replace human judgement in certain areas of healthcare. This is due to the increasing availability of healthcare data and rapid development of big data analytic methods. These technologies have been applied to image analysis in radiology, ophthalmology, and dermatology; to optimising treatment selection; the analysis of disease patterns and survival; and to the analysis of blood test results. The rate of change in this research area means that new approaches are continuously being developed.


Additional funding from the Cancer Policy Research Unit – led by Prof Stephen Duffy (QMUL)

Aims & objectives

We focus on which modalities of AI/ML for early detection and diagnosis of cancer are ready for possible implementation in primary care in the NHS.

Initial searches identified three applications of AI/ML technology that have relevance to primary care clinical practice; we will therefore conduct three separate systematic reviews:

  1. AI/ML-driven cancer risk prediction tools for the symptomatic population consulting in primary care (primarily using electronic health record data);
  2. AI/ML-driven image analysis algorithms applied to images of skin lesions to detect possible skin cancers
  3. AI/ML-driven image analysis algorithms applied to ultrasound images to facilitate point-of-care ultrasound in primary care (a technology that is more widely used in primary care in other European countries)

We will explore the AI/ML techniques that are being utilised and the clinical areas in which they are being employed. We will use the CanTest Framework to explore the stage of development that each technology has reached. We will also analyse and collate the diagnostic accuracy achieved, health-economic assessments that have been undertaken, and any implementation barriers that have been encountered.


We will follow systematic review methodology, and search Medline, Embase, Scopus and Web of Science for relevant published studies. Searches will start from the 1st January 2000, and extend until the date the search is performed for that section of the review. The search will be manually extended by: hand-searching of existing research archives and networks (e.g. the International Skin Imaging Collaboration, ArXiv, Google, Microsoft, IBM, Apple), and a parallel scoping review using Google searching to identify commercially-developed AI/ML technologies that do not have published data in academic journals.

Title and abstract screening will be completed by OJ with at least 15% checked by other authors. Data extraction will then be undertaken independently by at least two reviewers. Quality assessment of the included studies will utilise the QUADAS-2 Critical Appraisal Tools.

Outputs & impact

Initial searches suggest that studies in this research area are heterogeneous and of variable quality. We will therefore employ a narrative synthesis approach, alongside simple statistical analyses to provide an overview of the quantitative outcomes. We will perform meta-analyses if the data identified is thought to be suitable.

The wider impact of the study will be to provide an overview of the current state of AI/ML research on the early diagnosis of cancer in primary care settings, and to identify any technologies that are potentially ready for implementation in clinical practice. For technologies that are not yet ready for implementation but demonstrate potential, the review will identify the future steps needed to develop these technologies to the point where they will be ready for implementation.

Next steps

The findings will underpin further workstreams evaluating the diagnostic accuracy, acceptability and cost-effectiveness of AI/ML approaches for the early detection and diagnosis of cancer in primary care settings. The review may also identify next steps in the development of AI/ML technologies, and the evidence needed for implementation in primary care.


Back to Research