How AI has the potential to transform lung cancer screening in the NHS - Dr William Stephen Jones

While these developments seemingly mark a new milestone in public health, it is important to explore the details of these initiatives to understand their implications, practicalities, and potential pitfalls.

Lung cancer occurs when cells in the lung grow uncontrollably, damaging surrounding tissue and potentially spreading to other parts of the body. Lung cancer is currently the leading cause of cancer deaths worldwide, accounting for the highest mortality rates among both men and women. The lethality of lung cancer largely stems from its late detection, often when the cancer has spread extensively.

The challenge lies in identifying early lung cancer given its often indistinct and subtle early signs, and a promising solution to this is the implementation of a targeted screening program.

A screening program systematically and proactively tests for disease within a specific population. The NHS has established such programs, for example: breast cancer screening is offered to women aged between 50 and 71, with mammograms scheduled every three years; and cervical cancer screening is available to women aged between 25 and 64, with testing intervals from three to five years. The proposed lung cancer screening program targets individuals aged 55 to 74 with a history of smoking. These individuals would be invited to have a low-dose computed tomography (LDCT) scan, a type of x-ray, to check their lung health. If no issues are detected, then follow-up scans are offered every two years until they exceed the age threshold. The program is estimated to cost £270m annually at full operation, and the government expects to identify cancer in approximately 9,000 individuals each year, administering nearly a million scans. The anticipated outcome is earlier detection, leading to more effective treatment interventions.

This screening initiative represents a significant advancement in early detection and patient care, but there are several operational challenges. Perhaps the immediate concern is the increase in workload for radiology departments across the NHS, evaluating the additional images. These departments already face shortage of key personnel, a current shortfall of 33 per cent in workforce, expected to increase to 44 per cent by the year 2024, and are dealing with substantial backlogs, partly due to the pandemic. The added demand from the screening program will likely require immediate expansion of staff and resources to prevent further delays in diagnosis and treatment. Herein lies the value of AI to help with this process.

The promise of AI technologies is their ability to analyse medical images autonomously and rapidly, detecting cancers with accuracy comparable, or potentially surpassing, a human radiologist. Such advancements have scope to transform radiology, but the potential of AI extends well beyond use on medical images, with implications for all aspects of healthcare and society. However, the development of an AI technology for integration in the NHS is no trivial matter.

An AI technology functioning as a medical diagnostic test would be classified as a medical device. Consequently, it must adhere to the rigorous regulatory standards of medical devices, which demand extensive clinical evidence to demonstrate the AI’s accuracy and safety. Gathering the evidence can take several years, sometimes more than a decade, to collect as it involves running substantial clinical trials.

Before clinical trials can take place, the AI needs to be trained to detect the patterns of interest, requiring analysis of tens or even hundreds of thousands of lung cancer and non-cancer images. These images must be representative of the patient population that the AI will be used on, so the training data will likely need to come from NHS data.

Besides the regulatory and technical hurdles, AI diagnostics present ethical and legal challenges.

There are also data and privacy issues for patient data, training and ongoing education requirements for NHS staff to ensure effective use of the AI system. The system also needs regular monitoring and updates to maintain its performance, and it must be compatible with existing systems.

To ensure AI’s successful implementation for lung cancer screening, and beyond, these challenges need to be tackled head-on. If successful, AI will be transformative, leading to remarkable advances to medicine, for patient and healthcare system benefit.

Dr William Stephen Jones is director of research and lecturer in Data Science, Artificial Intelligence and Modelling (DAIM) at the University of Hull.